Executive Summary

• BLUF: 6G is evolving from a communications network into a distributed communications–sensing–computing–AI infrastructure capable of observing and interpreting physical space.
• The ITU has formally made Integrated Sensing and Communication — ISAC and AI and Communication — AIAC two of the six IMT-2030 usage scenarios.
• IMT-2030 targets positioning capabilities in the approximate 1–10 cm range and sensing functions including object detection, localization, velocity/angle estimation, imaging and mapping.
• This means future radio infrastructure can increasingly become a form of distributed radar, even when the target itself carries no communicating device.
• AI is not simply an application running over 6G: current international roadmaps increasingly treat AI inference, model distribution, network optimization and sensing interpretation as native network functions.
• Europe has committed at least €900 million of EU funding to the SNS Joint Undertaking for 2021–2027, matched by industry, while another €116 million for 20 projects was announced in March 2026.
• China moved in May 2026 to authorize 6 GHz spectrum for 6G technical trials and is targeting integrated communication-sensing-compute-intelligence capabilities toward 2030.
• The principal strategic risk is therefore no longer simply interception of communications: it is inference about people, objects, behaviour and environments from the radio environment itself.
• The supplied KIT claim concerning 197 volunteers and nearly 100% identification accuracy is not promoted to validated evidence in this report because its primary academic source falls outside the source hierarchy mandated here.
• Five-year assessment: the most probable 2031 outcome is not universal “6G surveillance”, but a heterogeneous world in which sensing-capable networks, AI-native infrastructure, regulation and geopolitical technology blocs develop simultaneously.

The Dawn of 6G: Transforming Communication with High-Speed Connectivity and Holographic Technologies

6G Will Turn Networks Into Sensors — And Power Into Sovereignty

The race for 6G is often described as the next contest for faster mobile connectivity. That interpretation is already obsolete. The architecture taking shape around IMT-2030 points toward networks that will communicate, sense, localize, compute and run artificial intelligence as integrated functions. Radio infrastructure will no longer be limited to transporting information: it will increasingly generate information about the physical environment itself. The economic prize therefore extends far beyond telecoms. It touches industrial automation, robotics, semiconductors, cloud computing, defence, space systems and data markets. The political risk is equally profound: infrastructure capable of locating objects, reconstructing environments and running distributed inference creates a new layer of surveillance and technological sovereignty. The strategic competition between Europe, the United States and China is becoming a contest over who will control that layer.

The Network Learns to See

The decisive document is Recommendation ITU-R M.2160, approved in November 2023. It defines the framework for IMT-2030 and introduces two capabilities that fundamentally distinguish 6G from previous generations: Integrated Sensing and Communication (ISAC) and Artificial Intelligence and Communication (AIAC). The same recommendation establishes a research target for positioning accuracy of 1–10 centimetres and explicitly identifies sensing functions including range, velocity and angle estimation, object detection, localization, imaging and mapping. It also describes AI-related capabilities such as distributed data processing, distributed learning, AI computing, model execution and inference. Framework and overall objectives of the future development of IMT for 2030 and beyond – International Telecommunication Union – November 2023.

That distinction matters. A conventional mobile network tries to overcome reflections, multipath propagation and movement because they distort communications. An ISAC network can use those same phenomena as information. Delay can reveal range; Doppler can indicate motion; phase changes can expose displacement; antenna arrays can estimate direction; multiple radio nodes can combine observations into a spatial representation. A base station therefore begins to acquire a second function: not only connecting devices, but measuring its surroundings.

On 17 March 2026, the ITU confirmed that Working Party 5D had completed the draft minimum performance requirements for IMT-2030. The framework contains 20 technical performance requirements, seven of them newly introduced for 6G, and formally preserves both AIAC and ISAC among the six usage scenarios. IMT-2030: Technical requirements for the 6G future – International Telecommunication Union – 17 March 2026.

From Cloud to Network Intelligence

The second break with the past concerns computing. AI today is predominantly cloud-centric: data travel from devices toward large computing centres, models execute remotely, and results return through the network. 6G is being designed for a different topology in which inference can migrate among devices, radio sites, metropolitan edge nodes, regional computing centres, satellites and central cloud infrastructure.

This is not merely an industry hypothesis. The ITU’s current IMT-2030 architectural work defines future networks as coordinating communications, sensing, computing, artificial intelligence and security. IMT-2030 networks – Considerations on framework – International Telecommunication Union – work item registered 4 March 2026.

The implication is that routing will increasingly become a computational decision. A future network may have to determine not only where a packet should go, but where an inference should run, which accelerator should execute it, whether raw data should move or the model should move instead, what latency is acceptable, what power budget exists and whether the computing node satisfies security requirements.

The United States has already institutionalized this shift. On 30 June 2026, the National Institute of Standards and Technology published the Communications Technology Laboratory’s 6G Communications Roadmap, based on six months of external stakeholder engagement and defining five research goals for the next five to seven years. The document places advanced communications, AI-native architectures, sensing, security and measurement science inside the same strategic research programme. Communications Technology Laboratory’s 6G Communications Roadmap – National Institute of Standards and Technology – 30 June 2026.

China Builds the Full Stack

China’s strategy is more explicitly integrated. On 8 May 2026, the Ministry of Industry and Information Technology authorized the IMT-2030 (6G) Promotion Group to use spectrum in the 6 GHz band for 6G technical trials in selected areas. The authorization is intended to support technology development, standardization and validation against ITU-defined 6G scenarios and performance indicators. Authorization of technical-trial frequencies for sixth-generation mobile communication systems – Ministry of Industry and Information Technology of the People’s Republic of China – 8 May 2026.

An even more revealing policy followed. The ministry issued its “AI + Information Communications” Innovation Development Implementation Opinion (2026–2028) on 3 June 2026, publishing it on 10 June 2026. It establishes concrete targets: by 2028, China intends to create more than 30 high-value application scenarios, achieve at least 75% coverage of a one-millisecond metropolitan computing-latency circle, and move information-communications networks toward higher levels of autonomous operation. By 2030, the ministry calls for a substantial strengthening of integrated communication-sensing-computing-intelligence capability. “AI + Information Communications” Innovation Development Implementation Opinion (2026–2028) – Ministry of Industry and Information Technology – 3 June 2026, published 10 June 2026, document MIIT Communications [2026] No. 121.

This architecture is strategically important because it joins spectrum policy, AI, edge computing and communications infrastructure before commercial 6G exists. The competitive unit is no longer the base station. It is the entire stack: spectrum, radio, chips, computing nodes, AI models, satellites, operating software and applications.

Europe’s Industrial Bet

Europe has chosen a different route: coordinated research, public-private investment and regulatory sovereignty. Council Regulation (EU) 2021/2085, adopted in November 2021, established the Smart Networks and Services Joint Undertaking. The programme carries an EU contribution of €900 million for 2021–2027, to be matched by private members, producing a minimum programme budget of €1.8 billion. Smart Networks and Services Joint Undertaking – European Commission – programme established November 2021.

On 2 March 2026, the European Commission announced another €116 million for 20 research and innovation projects, taking the SNS JU portfolio to 100 projects and bringing EU public funding already committed under Horizon Europe to €630 million. The Commission also stated that a further €270 million was planned for 2026–2027. The projects include AI-native orchestration, hardware-accelerated radio access, serverless computing, industrial AI, security and sovereign telco-cloud architectures. Europe to advance 6G innovation as 20 projects set to receive €116 million – European Commission – 2 March 2026.

Europe therefore possesses a substantial technological programme. Its strategic difficulty is converting research diversity into industrial scale. China can coordinate spectrum, infrastructure and industrial policy through a single state-directed framework. The United States combines world-leading cloud, semiconductor and AI ecosystems with federal spectrum and standards institutions. Europe must integrate telecom vendors, semiconductor assets, cloud infrastructure, national research programmes and twenty-seven regulatory jurisdictions without losing speed.

The Semiconductor Bottleneck

6G’s move toward distributed intelligence will also change semiconductor economics. Cloud AI concentrates compute in large data centres where power and cooling can be engineered around high-performance accelerators. Network-embedded AI requires inference at locations where electrical, thermal and physical constraints are much tighter: radio sites, industrial edges, vehicles and potentially non-terrestrial nodes.

The result will increase strategic importance not only for leading-edge logic, but for specialized accelerators, RF components, high-bandwidth memory, advanced packaging, chiplets, photonic interconnects and secure execution environments. A sensing-capable radio network that cannot process data locally merely exports its computational burden elsewhere; a sovereign network that depends entirely on foreign inference hardware or proprietary models remains sovereign only at the antenna.

This creates a new dependency chain. Telecom sovereignty can no longer be measured by who manufactures the radio equipment. It depends on who controls the chips, software stack, AI models, edge-computing architecture, cloud, sensing algorithms and updates that determine what the infrastructure is capable of perceiving.

The Privacy Boundary Moves

The most difficult question concerns people rather than machines. The ability to detect or localize an object does not automatically constitute biometric identification. Identification requires additional processing, discriminating features and usually comparison or correlation with reference information. Yet the legal boundary becomes increasingly important once radio-derived signals are used to infer persistent behavioural characteristics.

The EU Artificial Intelligence Act, Regulation (EU) 2024/1689, was signed on 13 June 2024, entered into force on 1 August 2024, and became generally applicable on 2 August 2026, subject to specific transitional provisions. Its definitions of biometric identification are technologically neutral and encompass physical, physiological and behavioural characteristics. Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence – European Parliament and Council – 13 June 2024.

This matters because switching from light to radio does not automatically move surveillance outside European law. If future radio systems are used to generate identifying behavioural features, their legal treatment will depend on the processing purpose, the data produced and the applicable provisions—not on whether the underlying sensor resembles a camera.

The deeper issue will be what happens below formal identification. Repeated radio observations can potentially generate occupancy histories, trajectories, behavioural patterns and environmental models. Combined with access-control systems, devices or other datasets, apparently non-identifying measurements can become far more revealing. The strategic asset is therefore not the waveform alone, but the ability to fuse radio-derived information with other data.

Civilian Networks, Military Relevance

The dual-use implications are unavoidable. Communications, sensing, localization, AI and autonomous systems are converging precisely as defence ministries are giving greater importance to the electromagnetic spectrum, unmanned systems, resilient command networks and distributed sensing.

An infrastructure capable of communications and sensing can potentially support industrial safety, autonomous vehicles or emergency response; the same technical primitives can contribute to drone detection, perimeter awareness, spectrum monitoring or distributed situational awareness. The consequence is not that commercial 6G networks automatically become military systems. It is that the technological boundary between communications infrastructure and sensing infrastructure becomes less absolute.

This has implications for deterrence and infrastructure protection. Telecommunications sites that materially support sensing or command functions could acquire greater strategic importance during crises. At the same time, terrestrial and non-terrestrial integration may provide resilience when conventional infrastructure is disrupted. The resulting security architecture will have to protect not only network availability, but the integrity of what the network believes about the physical world.

Data Becomes the New Spectrum

The most important economic change may ultimately occur after the radio measurement itself. Raw channel data have limited commercial value. AI can transform them into localization, occupancy, industrial movement, mobility patterns, asset trajectories or digital-twin updates. The valuable commodity becomes the inference.

That distinction could reshape competition between telecom operators, cloud platforms, equipment manufacturers and industrial companies. Whoever controls the sensing model controls what can be extracted from the physical environment. Whoever controls historical sensing data can improve that model. Whoever owns the interface through which those outputs are sold can build a platform around the network itself.

The 6G economy may therefore create a new category of data asset: information that users never consciously transmitted because it was generated from how radio waves interacted with the environment around them. The economic and regulatory battle will concern who may collect it, who may transform it, who may retain the inference and whether those outputs can be transferred across platforms or jurisdictions.

The Strategic Choice

The period to 2031 will not be defined by a single “6G launch”. It will be defined by decisions being made now about standards, spectrum, AI architectures, edge computing, semiconductor supply chains and sensing governance. The ITU has already placed sensing and AI inside the IMT-2030 framework; the United States has formalized a federal 6G research roadmap; China has allocated trial spectrum and connected communications directly to metropolitan computing and AI; Europe has committed a minimum public-private programme of €1.8 billion while bringing the AI Act into general application.

The strategic question is consequently larger than telecommunications. The next network generation will determine not merely how machines communicate, but where intelligence resides and who controls the infrastructure that converts the physical world into data. Countries that treat 6G as another mobile-network upgrade risk discovering too late that the real competition was over machine perception, computational sovereignty and control of the data generated by reality itself.

---

Drone War 2.0: AI, Mass Autonomy and Countermeasures

---

Navigational Index

Pillar I — From Telecommunications to Machine Perception

Radio sensing, ISAC, centimeter-class localization, distributed RF imaging, channel-state inference, sub-THz/mmWave evolution, edge intelligence, digital twins, robotics, autonomous systems and the transformation of the base station from communications infrastructure into a physical-world sensor.

Pillar II — AI-Native 6G and the New Computing Architecture

Distributed inference, network intelligence, semantic communications, AI-driven radio interfaces, sensing-data fusion, edge computing, autonomous network agents, terrestrial/non-terrestrial integration, cyber-physical systems, semiconductor implications and the migration from cloud-centric AI toward network-embedded intelligence.

Pillar III — Surveillance, Sovereignty and the 2026–2031 Strategic Contest

Privacy, biometric inference, regulatory boundaries, intelligence exploitation, counter-surveillance, military applications, spectrum sovereignty, technological blocs, Europe–United States–China competition, infrastructure dependency and the emerging political economy of radio-generated data.

---

Master Abstract

The central strategic error in analysing 6G is to regard it as a linear continuation of 5G measured primarily through throughput, latency and spectrum. The international standardisation trajectory now points toward something structurally different: a network that simultaneously communicates, computes, senses and increasingly interprets its surroundings. The ITU IMT-2030 Framework, approved in November 2023, establishes six usage scenarios that explicitly include Integrated Sensing and Communication and Artificial Intelligence and Communication alongside immersive communications, hyper-reliable low-latency communications, massive communications and ubiquitous connectivity. Its significance is architectural. Under ISAC, transmitted electromagnetic energy is no longer useful only when decoded by an intended receiver; reflections, delays, phase shifts, Doppler characteristics, angle of arrival and spatial propagation patterns become data describing the physical environment. ITU documentation explicitly anticipates detection, localization, movement tracking, posture and gesture recognition, pedestrian and vehicle detection, imaging, mapping and environmental sensing, including information about unconnected objects. The same framework identifies positioning in the approximate 1–10 cm range as a target capability, while technical feasibility work extends IMT consideration into frequencies above 100 GHz, where bandwidth, antenna aperture and propagation characteristics can support substantially richer spatial discrimination. This does not mean that every future 6G antenna becomes an omniscient camera, nor does it justify the literal claim that all humans can be identified with “100% accuracy”. It means something more consequential from an infrastructure perspective: telecommunications networks are acquiring a second informational channel. The first carries what users intentionally transmit; the second can be generated from what radio propagation reveals about their surroundings. Framework and overall objectives of the future development of IMT for 2030 and beyond – ITU – November 2023. ITU-R Recommendation M.2160 IMT towards 2030 and beyond – ITU Radiocommunication Sector – updated 2026. ITU IMT-2030 programme

The mechanism matters because it explains why camera-free sensing is technologically plausible without converting radio waves into conventional photography. A radio transmitter illuminates an environment electromagnetically. Walls, furniture, machines and human bodies absorb, scatter, diffract and reflect portions of the signal. Motion modifies the propagation channel; geometry affects multipath; moving surfaces produce Doppler information; antenna arrays provide angular information; wide bandwidth improves range resolution; and repeated observations create temporal structure. The resulting measurements can therefore become a multidimensional representation of an environment. Machine learning radically changes their informational value because models can learn relationships between radio-channel variations and physical states that would be difficult to encode analytically. The ITU’s November 2025 ISAC technical report already describes networks in which sensing nodes can be selected dynamically, sensing measurements can be processed across multiple sites, and cooperative measurements can be fused into final sensing results before those results are exposed to authorised applications or network functions. Crucially, the document also recognizes the sensitivity of sensing information and specifies authentication of sensing requests and sensing devices. This is the beginning of an entirely new security boundary. Protecting packet payloads is insufficient if useful information can be inferred from the propagation channel, sensing metadata or network-generated representations of physical space. Long before commercial 6G, the underlying direction is visible in WLAN research and standardisation: an ITU-hosted technical session on joint sensing and communication described uses of existing Wi-Fi infrastructure for presence detection, gesture recognition and person identification. Under the strict evidence protocol applied here, however, the specific Karlsruhe experiment cited in the prompt — 197 subjects with almost perfect identity recognition — remains outside the validated baseline because the underlying university publication is not among the source categories permitted by this report. The broader phenomenon, by contrast, is independently supported by ITU technical material. Considerations on integrated sensing and communication in IMT-2020 networks and beyond – ITU-T – November 2025. ITU-T YSTR.ISAC-fra

The second transformation is computational. AI-native 6G should not be understood merely as faster connectivity for accessing remote foundation models. The emerging architecture distributes intelligence across radios, edge nodes, core networks, satellites, devices and cloud infrastructure. ITU work initiated in 2026 describes IMT-2030 as coordinating communications, sensing, computing, AI and security, while NIST’s 2026 communications roadmap explicitly includes AI-native 6G applications and sensing among its strategic goals and is developing models, datasets and performance frameworks for joint communications and sensing. China is moving in a parallel but increasingly state-coordinated direction. On 8 May 2026, the Ministry of Industry and Information Technology authorised use of the 6 GHz band for 6G technical trials by the IMT-2030 promotion group. On 3 June 2026, MIIT launched a ministry–province collaborative 6G programme intended to generate indigenous technical solutions, new applications and terminal products by 2029, while its 2026–2028 “AI + Information Communications” programme targets by 2030 a substantial increase in integrated communications–sensing–computing–intelligence capability. These developments make the geopolitical dimension impossible to separate from the engineering dimension. Control over future networks means influence over spectrum policy, semiconductor platforms, radio hardware, AI accelerators, operating software, identity layers, sensing APIs, location infrastructure, satellite integration and the standards governing how observations of the physical world can be generated and exposed. Europe’s counter-strategy remains heavily research-driven: the Smart Networks and Services Joint Undertaking carries an EU contribution of €900 million for 2021–2027, intended to be matched by industry, while the Commission announced another €116 million across 20 projects in March 2026, bringing the SNS portfolio to 100 projects and stating that a further €270 million was planned for 2026–2027. The competitive object is therefore no longer “the fastest network”. It is the technological stack that determines who can transform ubiquitous electromagnetic infrastructure into a distributed perception-and-computation system. Smart Networks and Services Joint Undertaking – European Commission – current programme. European SNS Joint Undertaking 6G Innovation Development Ministry-Province Collaborative Pilot – MIIT – June 2026. MIIT 6G pilot programme

Telecom Cloud NFV Security: Privilege Escalation Paths Analysis

The privacy consequence is potentially historic because radio sensing changes the traditional relationship between observation, visibility and consent. Cameras are socially legible devices: people generally understand that a lens can observe them. Radio infrastructure is different. Wireless access points, base stations and connected devices already saturate houses, factories, offices, vehicles, streets and public spaces primarily because societies perceive them as communications infrastructure. If their emissions and channel measurements acquire progressively more powerful sensing functions, a capability transition may occur without an equally visible environmental transition. The regulatory implication is particularly significant in Europe. The EU AI Act defines biometric identification technologically rather than by reference to cameras and explicitly includes behavioural characteristics such as gait and posture within the biometric domain when used for identification. It also treats real-time remote biometric identification in publicly accessible spaces for law-enforcement purposes as particularly intrusive and subjects it to exceptionally restrictive conditions and enumerated exceptions. Whether a future RF-derived representation constitutes biometric data in any particular implementation would depend on what information is extracted, how it is processed and for what purpose; the legal outcome cannot simply be inferred from the sensing modality. Strategically, however, that technological neutrality matters enormously. Moving sensing from optics to radio does not automatically move it outside fundamental-rights law. The harder policy problem will involve systems that stop short of identity but continuously infer presence, movement, density, body geometry, behaviour, activity or occupancy. Such capabilities possess immense legitimate value: fall detection without wearable devices, industrial safety, robotic perception, vehicle awareness, smart factories, emergency response, digital twins, infrastructure monitoring and spectrum optimization. They simultaneously create intelligence and surveillance value. The policy contest of 2026–2031 will therefore concern who may sense, what may be inferred, how long measurements and derived features may persist, whether sensing requires an observable indicator, whether unconnected persons have enforceable rights, and whether sensing APIs can be technically restricted before raw measurements become behavioural intelligence. Regulation (EU) 2024/1689 – European Union – June 2024, consolidated 2026. EU Artificial Intelligence Act

The five-year intelligence outlook therefore has to reject two competing simplifications: H₁, that 6G is essentially “5G but faster”, and H₂, that it inevitably produces universal invisible surveillance. Four additional hypotheses better capture the strategic landscape: H₃, sensing becomes a commercially important but strongly permissioned network capability; H₄, different geopolitical blocs implement materially different sensing, privacy and exposure architectures; H₅, AI-native networking becomes more transformational than the radio-interface speed increase itself; and H₆, hybrid terrestrial/non-terrestrial sensing creates the deepest long-term military and intelligence consequences. Current evidence substantially weakens H₁ because ITU has already institutionalized sensing and AI as distinct IMT-2030 usage scenarios. H₂ remains technically conceivable in limited environments but is weakened by regulatory, physics, deployment-cost and standardisation constraints. H₃ and H₅ receive the strongest positive update from current ITU, NIST, EU and Chinese government programmes. H₄ rises because China is explicitly coordinating national and provincial 6G industrial development while the EU is embedding 6G inside a sovereignty-and-regulation framework rather than treating standards as purely commercial infrastructure. A 200,000-trial exploratory Monte Carlo model, using standardisation progress, device readiness, edge-AI maturity, regulatory friction, usable-spectrum availability and geopolitical fragmentation as uncertain drivers rather than historical frequencies, produces an analytical distribution in which geopolitical/architectural fragmentation is the modal 2031 state at approximately 33.0%, sensing-centric commercialization approximately 25.6%, an AI-native autonomous-network trajectory approximately 23.1%, a privacy-constrained trajectory approximately 9.0%, and a communications-first evolutionary trajectory approximately 9.3%. These are model outputs, not empirical forecasts, and their value lies in exposing sensitivities: reduce geopolitical fragmentation and regulatory divergence and the probability shifts toward globally interoperable sensing; increase restrictions on sensing-data exposure and the privacy-constrained branch expands; accelerate AI-at-the-edge and distributed inference and H₅ dominates. The strategic judgment is consequently stronger than any single probability: between 2026 and 2031, the central technological convergence will be the fusion of radio, radar-like sensing, computation and AI into infrastructure capable not merely of carrying information about reality, but of continuously generating machine-readable representations of reality itself.

China’s 2030 Execution Test: Growth, Security, Autonomy

Five-Year Bayesian Baseline — 20 August 2026

H₁ — 6G remains principally a communications upgrade: 9%
Evidence trend: declining.

H₂ — pervasive unrestricted RF surveillance becomes the dominant civilian model: 7%
Evidence trend: technologically possible in niches, institutionally constrained.

H₃ — ISAC becomes a major commercial and industrial capability under controlled exposure: 27%
Evidence trend: strongly increasing.

H₄ — geopolitical and regulatory fragmentation produces materially different 6G architectures: 33%
Evidence trend: increasing.

H₅ — AI-native networking becomes the primary differentiator of 6G: 24%
Evidence trend: strongly increasing.

The probabilities are structured analytic estimates and should be continuously updated as IMT-2030 candidate technologies, spectrum allocations, 3GPP specifications, national regulatory regimes and commercial chipsets become observable.

Pillar I — From Telecommunications to Machine Perception: How 6G Turns Radio Infrastructure into a Sensor of the Physical World

The fundamental technological discontinuity separating 6G from previous mobile generations is not a simple multiplication of bandwidth, spectral efficiency or peak data rate. It is the progressive conversion of the radio-access network into a distributed measurement infrastructure capable of extracting information from the physical environment through the same electromagnetic emissions used for communications. In the formal IMT-2030 framework, the International Telecommunication Union has already institutionalized this transition by establishing Integrated Sensing and Communication — ISAC as one of six usage scenarios for the future generation of mobile systems, alongside Artificial Intelligence and Communication, ubiquitous connectivity, immersive communication, massive communication and hyper-reliable low-latency communication. This is not merely semantic standardization. It changes the functional definition of a telecommunications network. A conventional base station transmits information to terminals and receives information from them; an ISAC-capable node can additionally evaluate propagation delay, phase evolution, Doppler shift, angle of arrival, multipath structure and signal-strength variations generated when radio waves interact with people, machines, vehicles, buildings and environmental surfaces. The physical environment therefore becomes part of the information system. ITU’s IMT-2030 work identifies positioning as a new performance dimension and frames future networks around capabilities that include sensing, localization and environmental awareness; the February 2026 technical-performance work establishes twenty minimum performance requirements for candidate IMT-2030 radio interfaces and explicitly retains ISAC as a native 6G scenario. Framework and overall objectives of the future development of IMT for 2030 and beyond – International Telecommunication Union – November 2023. IMT-2030: Technical requirements for the 6G future – International Telecommunication Union – March 2026.

The physics underlying this transformation is straightforward in principle but extraordinarily demanding in implementation. Any electromagnetic transmission propagating through a complex environment is altered by absorption, reflection, diffraction, scattering and interference. A human body is therefore not electromagnetically invisible: its geometry, composition, orientation and motion perturb the channel through which radio energy propagates. In conventional communications engineering, many of these effects have historically been treated as noise, fading or channel impairment that modulation schemes, coding systems, beamforming and equalization must overcome. In sensing-oriented communications, the same perturbations become information. Delay can reveal distance; angular information can contribute direction; Doppler components reveal relative motion; phase variations can provide sensitivity to extremely small displacements; multiple reflections create spatial signatures; and observations repeated across time expose movement trajectories. With sufficiently broad bandwidth, coordinated antenna arrays and sophisticated signal processing, the network can construct increasingly detailed representations of objects that were never active participants in the communications exchange. ITU’s November 2025 technical report on ISAC goes considerably beyond conceptual language: it defines a dedicated Sensing Function, sensing-service requests, sensing-node selection, processing of sensing measurements, multi-source and multi-node collaborative sensing, cloud-edge cooperation, historical-data storage, intelligent computing and controlled exposure of sensing results. In other words, the standardization trajectory is not merely toward “radar-like radio”. It is toward a service architecture in which environmental sensing becomes a schedulable network resource, analogous in operational logic to communications capacity. Considerations on integrated sensing and communication in IMT-2020 networks and beyond – ITU-T – November 2025.

Physical radio observable	Information potentially extracted	6G architectural significance
Propagation delay	Distance/range	Object localization, ranging, mapping
Doppler shift	Relative velocity/motion	Vehicle tracking, robotics, gesture or activity sensing
Angle of arrival/departure	Direction	Spatial localization, beam tracking
Phase variation	Micro-displacement	Fine motion, vibration, potentially physiological-motion sensing
Multipath structure	Environmental geometry	RF mapping, indoor localization, digital twins
Channel-state evolution	Movement/environmental change	Presence detection, activity inference
Multi-node correlation	Spatial consistency	Cooperative localization and distributed RF imaging
Historical sensing series	Behavioural/temporal patterns	Predictive analytics and AI-assisted classification

The phrase centimeter-class localization must nevertheless be treated analytically with precision. It does not mean that 6G networks will identify every individual with centimeter accuracy, nor does it imply omnidirectional penetration through arbitrary obstacles. Localization and identification are different technical problems. Localization estimates where an object is; identification attempts to establish what or who the object is. The former can be derived from geometric measurements such as time of arrival, time-difference of arrival, phase, directionality and multistatic geometry. The latter requires an additional inference layer: stable discriminative features, reference data, classification models and usually persistence across observations. The distinction matters because some public narratives collapse sensing, imaging and biometric identity into one capability and therefore exaggerate both technological maturity and inevitability. IMT-2030’s positioning objectives nevertheless represent a major shift because even localization without identity creates strategic value when combined with external systems. A radio network able to estimate location, trajectory and object class can be fused with access-control records, vehicle identifiers, logistics databases, robotic telemetry or existing visual sensors. Identity can therefore emerge from correlation even when radio sensing itself never produces a literal biometric match. This is the deeper intelligence problem: data fusion can transform non-identifying measurements into identifying context. Under a five-year horizon, the most consequential systems will therefore probably not be “RF cameras” operating alone but multisensor systems combining RF sensing with AI, maps, device telemetry, digital-twin models and historical behaviour. The technical boundary between telecommunications and situational awareness consequently dissolves progressively rather than instantaneously. ITU’s framework explicitly includes high-precision positioning, tracking, environmental sensing and collaborative sensing architectures, while a separate ITU work item addresses quality-of-service assurance for networks that possess awareness capabilities across both the physical and digital worlds. Study on quality of services assurance for integrated sensing and communication supported by IMT-2020 and beyond – ITU-T – 2026 work programme. Requirements and framework of integrated sensing and communication in IMT-2020 networks and beyond – ITU-T – February 2026 work item.

The next enabling layer is spectrum. Higher frequencies provide larger potential bandwidths and much smaller wavelengths, allowing more antenna elements to be integrated into limited physical apertures and potentially enabling finer spatial resolution, but they impose severe propagation penalties. Millimeter-wave technologies already illustrate this trade-off: narrower beams and wider channels improve spatial selectivity, yet penetration, blockage and atmospheric or material losses become increasingly important. Moving toward sub-THz and eventually selected ranges above 100 GHz increases this tension rather than eliminating it. The future therefore is unlikely to consist of one universal “6G frequency”. Instead, machine perception will probably depend on heterogeneous sensing: lower frequencies for coverage and resilience, mid-band spectrum for broad communication-sensing integration, millimeter-wave systems for higher spatial precision and sub-THz links for very high-capacity or short-range specialized use cases. The ITU’s IMT-2030 development process is explicitly extending propagation and channel-model work to include near-field effects, spatial non-stationarity and sensing-specific channel representations, which is strategically significant because conventional far-field abstractions become insufficient when arrays grow physically large relative to wavelength and when the network must estimate geometry rather than simply deliver bits. A base station using a large antenna array can exploit directional beams for communications, but the same spatial selectivity can also become part of a sensing architecture. Thus the antenna ceases to be merely a transceiver component and becomes part of a measurement aperture. The 2026 IMT-2030 evaluation preparations reflect this convergence by integrating sensing-specific channel models into the candidate-technology evaluation process rather than treating sensing as an application running independently above the radio layer. IMT towards 2030 and beyond – International Telecommunication Union – 2026.

The architecture becomes significantly more powerful once sensing is distributed across multiple nodes. A single radio link contains ambiguities: occlusion, multipath, interference and environmental change can degrade confidence. Multiple geographically separated transmitting and receiving nodes can observe the same environment from different geometries, generating a form of network-scale multistatic sensing. The ITU’s ISAC framework explicitly anticipates multi-mode, multi-source and multi-node collaborative sensing, dynamic sensing-node selection and processing that combines measurements from different sites. This is the bridge from “a base station that senses” to “a network that perceives”. In a factory, for example, several access points could observe the movement of autonomous mobile robots, workers and machinery; network nodes could fuse their measurements; an edge-compute system could reconcile them with a production digital twin; and the resulting state estimate could be delivered back to robots or industrial controllers. The same architecture could support airports, ports, logistics facilities, highways or dense urban infrastructure. Crucially, this topology creates new concepts of sensing coverage analogous to communications coverage. Operators may eventually have to reason about detection probability, localization geometry, sensing continuity, sensing interference and uncertainty boundaries just as they currently optimize signal quality and handover. Sensing requests could also be application-specific: one service may require rough occupancy, another precise motion trajectories, another high-confidence object localization. ITU’s architecture provides for a sensing-service requester, authentication, selection of suitable sensing resources, collection, processing and exposure of results, which implies a future service economy around “sensing as a network capability.” The security consequence is equally profound: authorization must govern not only who may transmit or consume data but who may ask the network to observe physical reality. Considerations on integrated sensing and communication in IMT-2020 networks and beyond – ITU-T – November 2025.

The decisive accelerator is edge intelligence. Raw sensing measurements are often too voluminous, noisy, environment-dependent and semantically weak to be useful without contextual processing. Artificial intelligence can perform denoising, classification, association, trajectory estimation, anomaly detection and multimodal fusion close to the point of observation. The strategic implication is that 6G sensing and AI development are mutually reinforcing: sensing provides real-world data for models, while AI converts electromagnetic measurements into higher-level meaning. The United States has formally entered this convergence through the NIST Communications Technology Laboratory 6G Communications Roadmap, published on 30 June 2026 after stakeholder engagement and designed to guide NIST’s 6G research investments over the next five to seven years. NIST explicitly frames advanced communications around performance, security, spectrum management and data privacy rather than viewing 6G as a single throughput race. China is pursuing an even more explicit communications-compute-sensing-intelligence fusion model. Its 2026–2028 “AI + Information Communications” Innovation Development Implementation Opinion requires integration between AI and 5G-Advanced/6G, research into AI-driven radio architectures, intelligent network equipment and edge inference, and states that by 2030 integrated communication-sensing-computing-intelligence service capability should be substantially strengthened. The same document calls for inference resources at communications base stations, core-network equipment, routers and optical systems, and identifies cloud-network-edge-device collaborative inference as a development direction. This is therefore no longer simply an industry forecast; it has entered state technology planning. Communications Technology Laboratory’s 6G Communications Roadmap – National Institute of Standards and Technology – June 2026. “Artificial Intelligence + Information Communications” Innovation Development Implementation Opinion 2026–2028 – Ministry of Industry and Information Technology of China – 2026.

2026–2031 capability layer	2026 baseline	Most plausible 2031 state	Principal constraint
ISAC architecture	Standardization and trials	Integrated into selected industrial/private networks	Standards maturity
Centimeter-class localization	Advanced research / defined target capability	Operational in controlled environments	Multipath, geometry, calibration
Distributed RF sensing	Early collaborative frameworks	Multi-node sensing in factories, mobility and critical infrastructure	Infrastructure synchronization
Edge AI	Rapidly expanding	Persistent local inference close to RAN	Energy and accelerator cost
RF digital twins	Experimental	Production-grade in high-value verticals	Model fidelity
Robotics integration	5G-A/private-network stage	Closed-loop perception–action systems	Safety validation
Sub-THz	Research and prototypes	Selective high-capacity/high-resolution deployments	Propagation and hardware
Network sensing exposure	Emerging architectural concept	Permissioned APIs in selected deployments	Privacy and cybersecurity

The digital-twin dimension is where the economic impact may become larger than the telecommunications market itself. A classical digital twin depends on explicit sensors: cameras, LiDAR, machine telemetry, GNSS, inertial measurements, industrial controllers and human-entered operational data. ISAC potentially turns the communications fabric into an additional always-present measurement plane. Consider a manufacturing site in which fixed radio nodes continuously observe space while connecting robots and machinery. A production digital twin could ingest radio-derived occupancy, motion and location estimates alongside machine-control data, producing a more continuously updated representation of physical operations. Autonomous robots would benefit because localization and environmental awareness would no longer depend exclusively on onboard sensing. Network infrastructure could contribute external perception, reducing blind spots and creating redundancy. The same model extends to connected vehicles and low-altitude aerial systems, where cooperative sensing from infrastructure may supplement onboard sensors. China has already linked its communications strategy to embodied intelligence, transportation, low-altitude economy and manufacturing applications through state planning for integrated edge inference. Europe’s public investment strategy takes a different institutional form but points toward a similar intersection between 6G and vertical-industry transformation. The European Smart Networks and Services Joint Undertaking has an EU contribution of €900 million for 2021–2027, intended to be matched by industry for a total budget of at least €1.8 billion, and its mission explicitly includes edge and cloud service provisioning, new devices and European technology sovereignty. In March 2026, the European Commission announced another €116 million for 20 6G projects, bringing the portfolio to 100 projects and identifying a further €270 million for planned 2026–2027 investments. National initiatives in Italy, France, Germany, Finland, Sweden, Spain, Ireland and the Netherlands are now being tracked together because 6G has become an industrial-policy object, not merely a telecommunications standard. The Smart Networks and Services Joint Undertaking – European Commission – 2026. Europe to advance 6G innovation as 20 projects set to receive €116 million – European Commission – March 2026. National 6G research initiatives across Europe – European Commission – May 2026.

Geopolitically, machine-perception infrastructure introduces an additional layer of technological sovereignty. China, the European Union, the United States and Russia are not pursuing identical 6G development models, and the asymmetry matters. China’s official programme integrates radio, AI, compute infrastructure and sensing into a coordinated industrial strategy with explicit 2030 objectives. The European Union combines research financing, standard-setting ambitions, industrial sovereignty and privacy/security governance. NIST’s roadmap positions the United States around metrology, interoperability, trusted performance evaluation, spectrum and advanced communications research. Russian government planning, while materially less detailed in publicly accessible 6G sensing documentation under the evidentiary rules used here, explicitly requires research for the creation and introduction of sixth-generation communication technologies and standards in the Russian Federation and includes future telecommunications technology within the national Data Economy and Digital Transformation programme. That difference in publicly documented maturity is analytically meaningful. It does not prove a Russian technological deficit in every 6G subsystem, but it indicates that the verified state documentation currently provides less detailed visibility into ISAC and AI-native network integration than the Chinese or international material. Government of the Russian Federation telecommunications-development planning – Government of Russia – 2026. National Project “Data Economy and Digital Transformation of the State” – Government of the Russian Federation – current programme. This divergence raises the probability that 6G becomes technologically fragmented. Different blocs may implement different policies governing sensing activation, data retention, edge inference, lawful access, export controls, trusted suppliers and interoperability, producing infrastructure in which the same radio standard supports materially different surveillance, industrial and security capabilities depending on jurisdiction.

The intelligence and cybersecurity consequences are larger than traditional eavesdropping because the attack surface expands from communications content into physical-world inference. A hostile actor does not necessarily need to decrypt application payloads if channel measurements, sensing metadata or improperly exposed sensing results reveal occupancy, movement or operational patterns. Conversely, an attacker could manipulate the sensing layer rather than steal data: spoofed reflections, malicious transmissions, adversarial waveform effects, compromised sensing nodes or poisoned historical observations could cause an autonomous system to form an incorrect representation of reality. That transforms wireless cybersecurity into a cyber-physical problem. A future warehouse robot could be misdirected because the network’s spatial model is corrupted; a smart factory might misclassify human presence; a security system could generate false objects or miss real ones; a communications network optimizing itself from sensing feedback could make poor resource decisions after adversarial manipulation. ITU’s decision to include authentication of sensing requests and sensing devices is therefore strategically important, but authentication alone cannot solve integrity, provenance, uncertainty and adversarial-machine-learning problems. The architecture will need confidence scoring, multi-node corroboration, anomaly detection, provenance tracking, isolation of compromised observers and explicit policies governing sensing-result exposure. At the state-security level, the same dual-use nature becomes unavoidable. Infrastructure designed for mobility or industrial automation can theoretically contribute to perimeter awareness, drone detection or facility monitoring. The analytical boundary between civilian telecommunications infrastructure and distributed situational-awareness infrastructure therefore becomes increasingly porous. This does not mean civilian 6G networks automatically become military sensors; it means the technical primitives required for sensing, multi-node fusion and AI-assisted interpretation can serve both domains, making governance and architecture more consequential than nomenclature.

Shadow dimension	Opportunity	Strategic risk	2031 warning indicator
Cyber norms	Authentication and sensing authorization standards	Unauthorized environmental observation	Unrestricted sensing-result APIs
Intelligence	Passive situational awareness	Infrastructure repurposed for surveillance	Persistent cross-site RF identity correlation
Liquidity/capital	New industrial 6G platforms	Capital concentration around proprietary sensing stacks	Vertical lock-in by network vendors
Supply chain	New RF front ends, AI accelerators, arrays	Semiconductor and trusted-vendor dependency	Export controls on sensing-capable components
Military-civil fusion	Resilient sensing and autonomous systems	Dual-use ambiguity	Commercial infrastructure integrated into defence C2
Privacy	Non-camera safety applications	Invisible observation	Persistent human classification without explicit device participation

A structured Analysis of Competing Hypotheses produces five plausible 2031 states for Pillar I. H₁ assumes communications performance remains dominant and sensing stays marginal; current ITU standardization materially weakens this hypothesis because ISAC is already a formal IMT-2030 scenario. H₂ assumes highly capable RF sensing develops but remains restricted mainly to factories, vehicles and controlled private networks; this remains highly plausible because those environments offer predictable geometry, economic return and clearer authorization. H₃ assumes sensing becomes a widespread public-network service exposed through regulated APIs; probability is increasing but depends heavily on privacy governance, standard maturity and operator economics. H₄ assumes sensing becomes deeply coupled to AI-native autonomous systems and digital twins, producing the largest industrial transformation; the convergence of ITU ISAC architecture, China’s explicit integrated communication-sensing-computing-intelligence policy and European vertical-industry investment strongly supports this path. H₅ assumes geopolitical fragmentation creates different national sensing architectures, data-exposure rules and vendor ecosystems; current evidence also raises this probability because technology sovereignty has become explicit in both Chinese and European policy and because Russia is prioritizing domestic telecom technology development. An illustrative Bayesian estimate for August 2026 assigns 12% to H₁, 25% to H₂, 22% to H₃, 25% to H₄ and 16% to H₅. By 2031, after applying scenario drivers for standards completion, AI-edge maturity, semiconductor availability, privacy regulation and geopolitical fragmentation, an analytical Monte Carlo model with 250,000 trials shifts the expected distribution to approximately 7% H₁, 22% H₂, 25% H₃, 28% H₄ and 18% H₅. These numbers are model outputs rather than empirical forecasts. Their principal value is to make the intelligence judgment explicit: the dominant uncertainty is no longer whether radio sensing becomes technically feasible, but where it is deployed, who controls its outputs, how AI interprets them and whether public or private networks expose those capabilities as standardized services.

The five-year trajectory should therefore be read as a staged transition rather than a single commercial launch. During 2026–2027, the decisive activity is standards definition, experimental platforms, channel models, hardware prototypes, edge-AI integration and high-value industrial trials. During 2028, architecture will increasingly converge around defined sensing functions, authorization, collaborative node selection and integrated AI processing, while regional regulatory approaches become more visible. Around 2029, early pre-commercial systems are likely to demonstrate closed-loop sensing and action in factories, mobility, robotics and controlled infrastructure, particularly where replacing or augmenting dedicated sensors creates an economic advantage. 2030 is the critical convergence point because international IMT-2030 development, European industrial programmes and China’s national targets are all structured around the end of the decade. By 2031, the most credible outcome is not a world blanketed by omniscient radio vision but a layered ecosystem: conventional communications remain ubiquitous; higher-frequency and high-density deployments provide localized precision sensing; private industrial networks become the leading proving ground for persistent ISAC; edge AI converts radio measurements into operational semantics; and digital twins become an important consumer of network-derived perception. The transformation of the base station will therefore be gradual but fundamental. Its historical role was to connect terminals. Its emerging role is to become a networked physical-world observer whose radio aperture, compute resources and AI stack participate in a distributed perception system. The strategic question for the 2030s will not simply be “how fast is the network?” It will be “what can the network infer about the world around it, with what certainty, under whose authority, and for whose benefit?”

Five-Year Indicator Matrix: 2026–2031

Indicator	2026	2027	2028	2029	2030–2031
IMT-2030 sensing standards	Framework/requirements	Evaluation refinement	Architectural consolidation	Pre-commercial alignment	Candidate deployments
Collaborative ISAC	Research/technical reports	Multi-node trials	Vertical prototypes	Integrated platforms	Operational niches
Edge sensing AI	Early network integration	Accelerated deployment	Specialized models	Closed-loop control	Native service layer
RF digital twins	Experimental	Industrial trials	Multi-sensor fusion	Production adoption	Broader vertical use
Sub-THz sensing	Research	Device improvement	Prototype systems	Selective trials	Specialized deployment
Public-network sensing APIs	Conceptual/early	Governance definition	Controlled trials	Early exposure models	Jurisdiction-dependent
Privacy/cyber rules	Fragmented	Policy formation	Standards interaction	Enforcement architecture	Regional divergence
Robotics dependency on network sensing	Limited	Assisted perception	Multi-sensor integration	Industrial scaling	Selective autonomy

Pillar II — AI-Native 6G and the New Computing Architecture: From Cloud-Centric AI to Network-Embedded Intelligence

The most consequential architectural change associated with 6G is that artificial intelligence is moving from being an external workload transported by the telecommunications network to becoming an internal mechanism through which the network itself is designed, controlled, optimized and eventually operated. This distinction separates an “AI-enabled network” from an AI-native network. In the first model, algorithms are added to existing network-management systems to predict congestion, optimize energy consumption or automate fault detection. In the second, intelligence is distributed across the architecture: radio-access functions, edge-computing nodes, transport networks, core functions, satellites, devices and cloud infrastructure continuously exchange state information, models, intentions and inference results, allowing resource allocation to be recalculated dynamically according to application objectives. The International Telecommunication Union now explicitly defines IMT-2030 networks as systems coordinating communications, sensing, computing, AI and security, while its Study Group 13 work programme identifies AI/ML, coordination of networking and computing, autonomous networking, semantic-aware networking, digital twins and ISAC as architectural capabilities to be incorporated rather than isolated applications. Even the radio-interface evaluation process reflects this shift: the IMT-2030 framework contains Artificial Intelligence and Communication — AIAC as one of six formal usage scenarios, while the February 2026 technical requirements establish twenty minimum performance criteria for candidate systems and the June 2026 evaluation framework introduces seven test environments. The implication is structural: future network performance cannot be evaluated solely as throughput delivered through a passive pipe. Networks must increasingly be assessed according to how efficiently communications and computational resources support intelligence. IMT-2030: Technical requirements for the 6G future – International Telecommunication Union – March 2026. IMT-2030 networks – Considerations on framework – International Telecommunication Union – March 2026. ITU IMT‑2030 technical requirements

The computing consequence is equally fundamental because centralized cloud inference becomes economically and physically inadequate for a growing class of 6G applications. A conventional cloud-centric AI pipeline transports sensor or application data from the device through the access network, aggregation network and core toward a regional or hyperscale data centre; inference occurs there, and the result returns across essentially the same path. That architecture remains efficient for workloads where latency is flexible, context changes slowly and model size dominates over reaction time. It becomes problematic when millions of machines, robots, vehicles, cameras, RF sensors or autonomous agents generate continuously changing states requiring decisions within strict latency, bandwidth, reliability or privacy constraints. The alternative is a compute continuum: inference may execute wholly on-device, at the radio site, at a metropolitan edge, in a regional AI cluster, in a central cloud or across several of them simultaneously. ITU Recommendation Y.3402, approved in December 2025, already specifies management and orchestration for coordinating networking and heterogeneous computing resources. Its architecture requires computing-node registration, unified resource measurement, real-time monitoring of metrics including CPU/GPU utilization, delay and bandwidth, task migration, resource reservation, service scheduling and mechanisms for routing applications toward suitable computing capability. More revealing still, an ITU work item created for wide-area distributed inference explicitly states that traditional single-point inference faces limitations in performance, cost and scalability as LLM, AI-agent and embodied-intelligence workloads expand, and proposes distributed inference across geographically separated computational nodes. This effectively changes routing theory: future networks may no longer choose a path merely to reach a destination address; they may choose a path because a particular node possesses the accelerator, memory, model fragment, energy budget, trusted execution environment and latency profile necessary to execute a portion of the intelligence workflow. Coordination of networking and computing in IMT-2020 networks and beyond – Management and orchestration – ITU-T – December 2025. Requirements and framework of future networks for supporting wide-area distributed inference – ITU-T – March 2026.

AI execution tier	Approximate architectural location	Likely 6G role	Critical limiting factor	Data movement implication
Device AI	UE, robot, vehicle, sensor	Immediate inference, local privacy, perception	Power, memory, thermal envelope	Lowest network dependence
Deep edge	Base station / RAN compute	Radio optimization, sensing inference, robotics	Accelerator density, cooling	Ultra-short data path
Metro edge	Metropolitan compute node	Multi-user inference, model aggregation, digital twins	Resource contention	Regional traffic reduction
Regional AI fabric	Carrier or sovereign data centre	Larger model execution, orchestration	Network/compute scheduling	Intermediate aggregation
Hyperscale cloud	Central data centre	Training, large inference, global optimization	Latency, transport cost	Highest backhaul demand
NTN compute	Satellite, HAPS, gateway	Remote inference, orbital routing, coverage continuity	Radiation, energy, intermittent links	Delay-dependent distribution
Cooperative continuum	Combination of tiers	Model splitting, agent federation	Orchestration complexity	Dynamic computation routing

The technical architecture required to make such distributed inference practical is substantially more complex than simply placing GPUs in telecom facilities. A network-embedded AI system has to solve at least seven simultaneous scheduling problems: where the data resides, where the model resides, where compute is available, what latency the application can tolerate, how much energy execution consumes, what trust level the node satisfies and how much network capacity is required to move either data or model parameters. This creates a new optimization space in which communication and computation become substitutable resources. If moving 500 megabytes of sensor data toward a model is expensive, it may be preferable to move a compressed model toward the sensor; if the edge node lacks sufficient accelerator memory, an inference graph can be partitioned so that early layers execute locally while later layers execute at a regional node; if radio conditions deteriorate, the scheduler may relocate inference before a service-level violation occurs. ITU’s February 2026 Study Group 13 meeting shows how quickly this concept is entering formal architecture work: it initiated Y.OAMTI, covering orchestration for AI model training and inference; Y.FMSC-EI, covering edge intelligence in fixed-mobile-satellite convergence; and Y.FMSC-AI, covering AI in converged networks. The same meeting consented Y.3193, addressing distributed intelligence collaboration, and initiated work on edge-computing services specifically for AI workloads and observability of AI workloads in cloud systems. This collection of standards activity is more important than any individual specification because it reveals the emerging control plane: compute location, AI lifecycle, telemetry and workload orchestration are converging with network control. By 2030–2031, therefore, the operator’s scheduler could become conceptually closer to a distributed operating system than to today’s resource-management software. Executive summary of the SG13 meeting – International Telecommunication Union – February 2026.

The radio interface itself represents the next frontier of AI-native transformation. Historically, cellular physical layers have relied on algorithms whose behaviour is extensively engineered and mathematically bounded: channel estimation, coding, modulation, beam management, scheduling and interference coordination operate through known protocols and deterministic constraints. 6G AI-native radio introduces the possibility that machine-learning components perform parts of those functions or influence their parameters dynamically. This is much more consequential than using AI for network analytics because a model can enter the control loop at millisecond or sub-millisecond operational timescales. ITU’s future-technology work had already identified an evolutionary path toward an AI-native radio network architecture, including automated optimization, radio-resource adjustment, fault recovery, scheduling, network planning and energy optimization; it simultaneously identified unresolved challenges such as lack of bounded performance, explainability limitations, generalization uncertainty and interoperability. NIST’s June 2026 6G roadmap places AI-native 6G applications as its first strategic research goal and specifically proposes datasets and benchmarks for propagation, sensing, spectrum usage and related network environments, while identifying AI models for propagation, interference and sensing as potential standards contributions. That emphasis on datasets is crucial. AI-native radio cannot be trusted on the basis of model accuracy averaged over laboratory conditions; it must remain robust when confronted with environments absent from training data, adversarial interference, hardware drift, new antenna geometries or emergency operating conditions. A deterministic equalizer may degrade predictably; an AI component may fail discontinuously. Network engineering must therefore add uncertainty estimation, out-of-distribution detection, fallback algorithms, model version control, provenance and bounded-action policies to the radio stack. Future technology trends of terrestrial IMT systems towards 2030 and beyond – International Telecommunication Union – November 2022. Communications Technology Laboratory’s 6G Communications Roadmap – National Institute of Standards and Technology – June 2026.

Radio/network function	Classical architecture	AI-native 6G candidate	Failure mode introduced by AI
Channel estimation	Mathematical estimator	Learned channel representation	Out-of-distribution propagation
Beam management	Codebooks/rules	Predictive beam selection	Model drift / incorrect prediction
Scheduling	Optimization heuristics	Reinforcement-learning scheduler	Unstable or unfair policies
Interference management	Explicit measurements	Learned prediction	Adversarial/environmental mismatch
Fault management	Threshold alarms	Predictive diagnosis	False attribution
Energy optimization	Static/dynamic policies	AI closed-loop control	Performance-energy oscillation
Mobility management	Handover thresholds	Predictive trajectory-aware control	Biased or erroneous mobility model
Spectrum access	Rules/database coordination	AI-assisted dynamic allocation	Misclassification / harmful interference
Network slicing	Provisioned SLA templates	Intent-driven adaptive slices	Autonomous policy conflict
Sensing fusion	Dedicated algorithms	Multimodal learned fusion	Hallucinated or corrupted physical state

Semantic communications may produce an even deeper departure from traditional networking because it challenges the assumption that the purpose of the network is to reproduce every transmitted bit with extremely low error probability. The emerging semantic approach asks instead whether the receiver obtained the information required to complete a task. A robotic system does not always require an exact replication of every frame captured by another robot; it may need only the location, class and trajectory of an obstacle. An industrial controller may not need an entire sensor history; it may need a validated anomaly state. A language-agent interaction may prioritize intent or actionable meaning rather than token-perfect reproduction of intermediate representations. Under these conditions, communication efficiency can potentially be improved by transmitting task-relevant representations instead of raw data. This concept should not be oversold: semantic communications remain a research and standardization frontier rather than a mature replacement for conventional data transport. Nevertheless, ITU is developing a formal reference architecture for semantic-aware networking, scheduled for 2027, which defines machine- and human-shared semantic terms and syntax for representing, annotating, analysing and interpreting network and user-generated data. Additional February 2026 ITU work includes an architectural framework for knowledge-based semantic communication across public mobile networks and a semantic-communication pipeline using generative AI reconstruction. The strategic significance lies in the interaction with distributed inference. Once networks understand application semantics, scheduling decisions can incorporate task importance rather than packet characteristics alone: two traffic flows of equal bitrate might receive radically different treatment because one contains low-value redundancy while the other contains a safety-critical state update. This could produce goal-oriented networking, where application intent flows downward into radio, compute and routing decisions. Europe is explicitly funding this direction: the Commission’s 2026 SNS portfolio identifies MAGIC-6G as pursuing goal-oriented networking using semantic AI models. Requirements and reference architecture of semantic-aware networking in future networks – International Telecommunication Union – updated July 2026. Europe to advance 6G innovation as 20 projects receive €116 million – European Commission – March 2026.

The autonomous-agent layer is the point at which network intelligence moves from optimization toward delegated decision-making. An autonomous network agent can receive a high-level objective — reduce latency for a robotic production cell, restore resilience after an optical failure, minimize energy while maintaining an SLA, or establish connectivity for emergency-response traffic — and decompose that objective into actions across multiple network domains. Unlike conventional automation scripts, an agent can theoretically interpret telemetry, invoke tools, negotiate resources, compare options, execute changes, verify outcomes and revise its strategy. Formal standardization is already beginning to acknowledge this architecture. ITU Study Group 11 documented work in March and July 2026 on signalling and protocols for AI-agent communication networks, AI-enabled cross-domain networks, distributed data planes, digital-twin-supported networks and coordination of computing and networking across fixed, mobile and satellite convergence. ITU Study Group 13 separately initiated work on defining AI agents in future networks and AI Agent as a Service. The European Commission’s March 2026 portfolio goes further operationally: Agentic6G is described as developing autonomous multi-agent systems capable of self-organization, self-healing and secure service orchestration across the device–edge–cloud continuum, while 6G-OPTICON targets end-to-end AI-native orchestration. These projects are still research initiatives, but collectively they reveal the architectural direction. The decisive safety requirement is hierarchical authority. A network agent cannot be allowed unrestricted ability to modify routing, spectrum configuration, access control and compute placement simply because its predicted utility is high. Production systems will require action envelopes, privilege separation, explainable intent translation, change simulation in network digital twins, rollback capabilities, human override and cryptographically verifiable provenance. Otherwise, agentic networking converts software hallucination into infrastructure action, an error class much more consequential than incorrect text generation.

Autonomous-network control level	Decision authority	Example capability	2031 plausibility	Principal governance requirement
L₀ Manual	Human	Configuration command	Already universal	Operator discipline
L₁ Assisted	AI recommendation	Fault ranking	Mature	Validation
L₂ Conditional automation	AI executes bounded policy	Load balancing	High	Policy limits
L₃ Closed-loop domain autonomy	AI controls one domain	RAN optimization	Very high	Fallback + audit
L₄ Cross-domain agentic control	Agents coordinate RAN/core/cloud	SLA realization	Moderate-high	Privilege architecture
L₅ Intent-driven multi-agent network	High-level goals trigger autonomous orchestration	Self-reconfiguring network	Selective by 2031	Strong formal assurance

The terrestrial/non-terrestrial convergence adds another dimension because computation and intelligence will no longer exist exclusively in ground infrastructure. The future architecture can include terrestrial macro cells, local edge nodes, low-Earth-orbit satellite systems, high-altitude platforms, airborne relay systems and gateways, each presenting radically different delay, mobility, energy and compute characteristics. This means the orchestration plane must decide not only whether to route through satellite or terrestrial access but also whether inference should execute before, during or after traversal of the non-terrestrial segment. ITU’s February 2026 work formally includes fixed, mobile and satellite convergence, user-plane enhancement, edge intelligence and AI considerations, while a June 2026 ITU Journal study developed and testbed-evaluated an AI-driven orchestration architecture for integrated terrestrial-satellite 6G networks incorporating an enhanced Network Data Analytics Function, service-hosting environment, knowledge exposure and AI-agent framework. At European level, the SNS JU and European Space Agency signed a Memorandum of Intent in October 2025 specifically to deepen terrestrial/non-terrestrial 6G integration, while the 2026 EU portfolio includes 6G-NTN2 Nexus for terrestrial-satellite integration. This convergence has profound implications for resilience. An industrial or public-safety AI system may continue operating if terrestrial infrastructure fails by migrating selected functions toward satellite connectivity or remote compute, but such resilience creates new optimization problems because satellite capacity and latency are constrained and orbital compute has extreme power and thermal limitations. Future agents therefore require topology awareness across both compute and communications. A task may be split so that event detection executes locally, compressed semantic state traverses the satellite network and high-order reasoning executes remotely. The economic unit becomes neither “bit” nor “GPU second” alone, but a composite communication-compute-energy-latency transaction. AI-driven orchestration for integrated satellite-terrestrial 6G networks – International Telecommunication Union – June 2026. Smart Networks and Services Joint Undertaking and European Space Agency cooperation on next-generation connectivity – European Commission – October 2025.

The cyber-physical consequence emerges when network-embedded intelligence is connected directly to machines. 6G-driven AI architecture becomes especially transformative in robotics, autonomous vehicles, low-altitude aviation, logistics, smart factories and infrastructure automation because communication, sensing and computation form a closed perception-decision-action loop. Consider a factory robot whose onboard cameras and proprioceptive sensors are supplemented by network-level RF sensing. Local AI identifies immediate hazards; edge infrastructure fuses observations from multiple machines; a digital twin predicts congestion or collision; the network orchestrator allocates radio and compute resources; and an autonomous controller modifies trajectories. The entire loop may operate continuously. A communications outage therefore ceases to be merely a loss of connectivity; it can become a degradation of machine perception or control. Conversely, compromised telemetry can trigger physical actions. NIST’s July 2026 AI/ML Roadmap for Smart Manufacturing explicitly identifies industrial big-data analytics, advanced sensing, autonomous systems, digital twins, robotics, supply chains and trustworthy operation as central AI-development domains and highlights the continuing difficulty of integrating heterogeneous sensing and control systems. These requirements align precisely with the 6G architecture being developed internationally. The security model must therefore treat AI model integrity, inference provenance, communications integrity and physical-safety constraints as one continuous assurance problem. A digital signature proving that a network message originated from an authorized edge node does not prove the node’s AI inference is correct; a correct model does not guarantee its input sensing data were authentic; authentic sensing does not guarantee the downstream robotic action is safe. By 2031, credible industrial 6G architecture will therefore require multi-layer assurance spanning sensor provenance, model provenance, runtime monitoring, deterministic safety envelopes, redundant perception and controlled degradation modes. 2026 Roadmap on Artificial Intelligence and Machine Learning for Smart Manufacturing – National Institute of Standards and Technology – July 2026.

The semiconductor dimension is therefore inseparable from network architecture. Cloud-centric AI concentrates expensive accelerators in environments where hundreds of kilowatts or megawatts can be delivered and cooled efficiently. Network-embedded intelligence distributes compute across locations where power, cooling, space and cost are much more constrained. A base-station site cannot simply replicate a hyperscale AI rack. This creates demand for heterogeneous silicon: general CPUs for orchestration, network processors and DPUs for packet handling, specialized NPUs or ASICs for inference, RF transceivers and beamforming ICs, high-bandwidth memory, non-volatile memory for local model storage, photonic or high-speed die-to-die interconnects and advanced packaging capable of placing specialized chiplets within energy-efficient packages. The 2025 TSMC Annual Report, an audited corporate source allowed under this report’s evidence hierarchy, directly identifies AI accelerators, GPUs, ASICs, FPGAs, server processors and high-speed networking chips as products applicable to future 5G/6G infrastructure, AI and cloud systems. It further describes 2 nm N2, 3 nm and other advanced technologies, plus 3DFabric, SoIC and CoWoS packaging for heterogeneous integration, compute density, low latency and energy efficiency. The report explicitly links advanced packaging and high-performance chips to edge-AI applications and states that its SoIC platform targets improved interconnect bandwidth and power efficiency. These are not peripheral manufacturing details: in a distributed 6G system, inference performance increasingly depends on data movement between compute, memory and radio interfaces rather than transistor count alone. TSMC 2025 Annual Report – Taiwan Semiconductor Manufacturing Company – 2026.

The United States is treating these semiconductor bottlenecks as national infrastructure issues rather than isolated commercial engineering problems. The April 2026 CHIPS Research and Development Office Broad Agency Announcement explicitly identifies next-generation memory, heterogeneous integration, 3D packaging, chiplets, 3D stacking, 2.5D packaging, integrated photonic packages and ultrahigh-density interposers as priority technical areas, alongside AI-enabled system design and edge AI. It also identifies compute efficiency, novel memory stacks, AI devices for extreme environments and AI-driven autonomous agents as relevant semiconductor research domains. NIST separately defines advanced packaging as essential to AI because tightly integrating multiple dies increases functionality while reducing power and communication distance. This creates a strategic semiconductor hierarchy for 6G: leading-edge logic matters, but memory bandwidth, packaging, RF technology, photonics, power delivery and thermal management may determine whether AI-native base stations and edge nodes are economically deployable. The most advanced training accelerator is useless at a radio site if its power budget or cooling requirements exceed what the infrastructure can support. Conversely, a lower-precision accelerator optimized for radio inference, sensing fusion or small language/agent models may produce much higher network-level utility per joule. The likely 2031 market therefore shifts from a binary distinction between handset chips and data-centre chips toward a continuum of specialized compute nodes. This also widens geopolitical exposure: export controls or production constraints affecting advanced logic, HBM, substrate manufacturing, packaging equipment, photonics or RF front ends can constrain AI-native 6G independently of traditional telecom equipment availability. CHIPS Research and Development Office Broad Agency Announcement – NIST/US Department of Commerce – April 2026.

Semiconductor layer	Function inside AI-native 6G	Critical engineering variable	Strategic supply-chain exposure
Leading-edge CPU/GPU/NPU	Inference and orchestration	Performance per watt	Advanced-node foundry capacity
AI ASIC	Domain-specific inference	Efficiency / deterministic latency	Design ecosystem + foundry
HBM / advanced memory	Model weights and activations	Bandwidth / energy per bit	Memory concentration
Chiplets	Heterogeneous integration	Die-to-die bandwidth	Interface standards
2.5D/3D packaging	Accelerator-memory proximity	Yield / thermal density	Packaging capacity
RF front end	Radio access and sensing	Frequency / noise / linearity	Specialty processes
Beamforming IC	Massive arrays	Power/channel	RF semiconductor ecosystem
Silicon photonics	High-speed interconnect	pJ/bit	Packaging integration
DPU / network processor	Packet/model-flow orchestration	Throughput/watt	Networking silicon
Secure enclave	Trusted AI execution	Isolation / attestation	Security architecture
Non-volatile memory	Local model storage	Density / endurance	Process integration

China’s official strategy is unusually explicit and provides the strongest government-level evidence that communications and AI are being developed as one infrastructure system. The Ministry of Industry and Information Technology’s “AI + Information Communications” Innovation Development Implementation Opinion (2026–2028), issued on 3 June 2026 and published on 10 June, establishes a target that by 2028 AI and communications should form an initial mutually reinforcing development system, that communications networks should reach a higher degree of autonomy, and that more than 30 high-value application scenarios should be created. More strategically important is its metropolitan computing target: at least 75% coverage of a one-millisecond metropolitan computing latency circle by 2028. By 2030, MIIT calls for substantial breakthroughs in core AI-communications technologies and a major increase in integrated communication–sensing–computing–intelligence capability. The document specifically calls for AI-driven network architectures, intelligent mobile-radio interfaces, high-level autonomous networks, network-native intelligence, space-based computing networks and an “Internet of agents.” This is not simply a vision of AI applications consuming telecom bandwidth. It is an explicit state programme to create a geographically distributed compute fabric in which the network becomes part of China’s AI infrastructure. The policy also builds on a much larger existing network base: China’s previous national 5G scale-up programme targeted 70,000 industry virtual private networks and 5,000 edge-computing nodes by the end of 2027, demonstrating that the physical deployment substrate for distributed AI is being built before 6G arrives. “Artificial Intelligence + Information Communications” Innovation Development Implementation Opinion 2026–2028 – Ministry of Industry and Information Technology of the People’s Republic of China – June 2026. 5G Scale Application “Sailing” Action Upgrade Plan – Ministry of Industry and Information Technology – 2024.

Verified Chinese target	Deadline	Strategic implication
More than 30 high-value AI–communications scenarios	2028	Accelerated commercial verticalization
Metropolitan 1 ms compute circle ≥75% coverage	2028	Geographic distribution of inference
Higher-level autonomous communications networks	2028	Network-agent foundation
AI-driven air-interface research	2026–2028	Intelligence moves into radio
Space-based computing-network research	2026–2028	NTN + compute convergence
“Internet of agents” development	2026–2028	Agent-to-agent communications layer
Major increase in integrated communication-sensing-compute-intelligence capability	2030	Full convergence objective
70,000 5G industry virtual networks	2027 target	Private-network substrate
5,000 edge-computing nodes	2027 target	Pre-6G distributed compute base

Europe is converging on a similar technical architecture but through a markedly different political-economic model based on federated research, public-private financing, standardization and technological sovereignty. The Smart Networks and Services Joint Undertaking has an EU contribution of €900 million for 2021–2027, intended to be matched by private members for a minimum programme size of €1.8 billion. By March 2026, the Commission reported €630 million of EU public funding already committed, announced another €116 million across 20 projects, and stated that nearly 80% of SNS JU projects embed AI/ML as core components. The programme had generated more than 1,000 standardization contributions and 98 patents, and a further €270 million was planned across 2026–2027, including a 2027 flagship call exceeding €230 million. The project mix is architecture-specific rather than generically “6G”: Agentic6G targets autonomous multi-agent orchestration across device-edge-cloud; 6G-OPTICON targets end-to-end AI-native orchestration; MAGIC-6G targets semantic goal-oriented networking; 6G-NTN2 Nexus targets terrestrial-satellite convergence. Earlier EU projects also include 6G-CLOUD, designed around an AI-native cloud continuum, and 6G-INTENSE, focused on intent-driven native AI integrating compute-network abstraction and sensing at the deep edge. The Commission also identifies national programmes across Italy, France, Germany, Finland, Sweden, Spain, Ireland and the Netherlands, showing that Europe’s strategic challenge is not absence of technical ambition but coordinating fragmented national and industrial ecosystems quickly enough to create scale. Smart Networks and Services Joint Undertaking – European Commission – current programme. Europe to advance 6G innovation – European Commission – March 2026. National 6G research initiatives across Europe – European Commission – May 2026.

Russia presents a different trajectory under the strict evidence framework applied here. I do not find an equally detailed, currently verifiable Russian government architecture document that maps AI-native 6G, distributed inference, semantic communication and autonomous RAN control with the same specificity as MIIT or ITU documents; claiming one would violate the evidentiary rule. What can be verified is the construction of the enabling compute-sovereignty layer. On 16 March 2026, the Russian government announced a roadmap for high-performance computing, AI algorithms, grid technologies and supercomputer infrastructure and referenced a national objective of increasing aggregate domestic supercomputer capacity tenfold by 2030. Russian strategic policy also defines AI, communications technology, microelectronics, photonics, sensors, robotics and advanced mobile networks as priority technological areas, while the broader national AI strategy runs to 2030. The implication should therefore be framed cautiously: Russia is visibly prioritizing sovereign compute and domestic technology capacity, but the available primary sources do not justify assigning it a publicly documented AI-native 6G architecture equivalent to China’s “通感算智” programme. This asymmetry itself is an intelligence indicator. By 2031, the global architecture may divide not only between different telecom vendors but between ecosystems possessing deep integration of advanced edge silicon, distributed inference and agentic networking and those relying more heavily on centralized compute or domestically constrained hardware stacks. Roadmap for High-Performance Computing and Supercomputer Infrastructure – Government of the Russian Federation – March 2026. National Strategy for the Development of Artificial Intelligence to 2030 – Government of the Russian Federation – updated framework.

The cyber-security transformation follows directly from this architecture. In conventional networks, security boundaries can be approximately decomposed into device trust, transport confidentiality, core-network security and application security. In an AI-native network, additional trust surfaces appear around training data, inference data, model weights, model provenance, agent identity, computational placement, digital-twin state, sensing inputs and autonomous actions. ITU’s 2026 work on technical security controls for IMT-2030 explicitly identifies native AI, cloud-native and edge-native architectures, slicing, ISAC and heterogeneous multi-stakeholder environments as factors expanding the attack surface and changing trust boundaries. An attacker capable of compromising the model registry could distribute malicious network-optimization models without compromising base-station firmware. A compromised edge accelerator could return manipulated inference while packets remain cryptographically valid. A rogue autonomous agent could use legitimate APIs to reconfigure topology. Poisoned sensing data could cause a digital twin to represent a fictitious physical state. A model inversion attack could expose sensitive local observations, while federated-learning updates could themselves leak information. An NTN segment adds additional administrative domains and potentially intermittent connectivity, complicating attestation and revocation. The resulting security architecture must therefore adopt zero-trust AI execution: authenticated agents, signed models, remote attestation, confidential computation where feasible, fine-grained authorization, policy-constrained tool use, deterministic fallback, continuous model-health monitoring and immutable audit trails. Network intelligence cannot be secure merely because the underlying network is encrypted. The AI control plane itself becomes critical infrastructure. Technical security controls for IMT-2030 networks – International Telecommunication Union – June 2026 work programme.

New AI-native asset	Representative attack	Operational consequence	Required control
Model weights	Malicious replacement	Incorrect network decisions	Signing + attestation
Training dataset	Poisoning	Systemic behavioural bias	Provenance + validation
Inference input	Adversarial manipulation	Incorrect radio/control output	Robustness screening
AI agent identity	Impersonation	Unauthorized orchestration	Strong machine identity
Agent tool privileges	Prompt/tool abuse	Infrastructure reconfiguration	Least privilege
Digital twin	State corruption	Incorrect planning	Multi-source verification
Edge compute node	Runtime compromise	False inference	Trusted execution
Model-update channel	Supply-chain attack	Fleet-wide compromise	Signed deployment chain
Federated updates	Gradient poisoning	Distributed degradation	Byzantine-robust aggregation
Semantic representation	Meaning distortion	Task failure despite valid packets	Semantic integrity checks

The five-year forecast therefore turns on competing architectural hypotheses rather than on one binary “6G succeeds/fails” judgment. H₁ — Cloud persistence assumes the majority of meaningful AI remains centralized and 6G principally provides faster transport. H₂ — Hybrid distributed intelligence assumes edge execution expands but cloud remains dominant for training and large models. H₃ — AI-native agentic network assumes model orchestration, distributed inference and autonomous control become fundamental network functions. H₄ — Sovereign fragmentation assumes technology blocs create materially different compute, AI and networking stacks. H₅ — Constraint-dominated evolution assumes energy, security, semiconductor or trust problems significantly delay autonomy. I executed an exploratory 250,000-trial Monte Carlo scenario model using nine uncertain drivers — standardization maturity, edge-AI maturity, compute availability, semantic-network maturity, agentic-control maturity, NTN integration, energy efficiency, security readiness and geopolitical fragmentation — represented through bounded probability distributions rather than historical frequencies. Under an August 2026 baseline, the resulting scenario frequencies were approximately 21.8% H₁, 21.0% H₂, 26.8% H₃, 14.2% H₄ and 16.2% H₅. These values are analytical outputs, not observations and not probability statements endorsed by any cited institution. More important than the point estimates are the sensitivities: H₃ rises sharply when edge compute, agent reliability and semantic networking mature together; H₅ rises when energy and trust lag; H₄ rises disproportionately when advanced-chip supply and standards fragmentation correlate. The Bayesian update from the 2026 evidence therefore increases H₃ relative to a 2023 baseline because autonomous-agent, distributed-inference and compute-network coordination have migrated from research rhetoric into formal ITU work items and funded national programmes. H₁ remains material because cloud economics and foundation-model scale continue to favour centralization, but the evidence no longer supports treating central cloud as the exclusive locus of intelligence.

2026–2031 AI-Native 6G Technology Maturity Matrix

Vector	2026 verified state	2027–2028 inflection	2029–2030 probable state	2031 analytical judgment
Distributed inference	Standards work underway	Wide-area orchestration frameworks	Production vertical pilots	High adoption in controlled networks
AI-RAN	Research + benchmarking	Model lifecycle standardization	Selective closed-loop functions	Broad assistance, bounded autonomy
Semantic networking	Reference architecture under study	Initial formal specifications	Goal-oriented pilots	Selective application-specific use
Autonomous agents	Definitions/protocol work	Domain agents	Multi-domain trials	Significant but constrained
Edge AI	Mature precursor infrastructure	Rapid accelerator deployment	Deep-edge inference	Foundational 6G capability
AI workload orchestration	Standards initiated	Compute-network scheduling	Dynamic model placement	Core service
Digital network twins	Existing research architecture	Integrated testing	Agent validation/simulation	Operational control component
Terrestrial/NTN AI	Active integration programmes	Cooperative orchestration	Initial commercial convergence	Important coverage/resilience layer
Semiconductor heterogeneity	Advanced packaging expansion	Chiplet/AI-edge acceleration	Specialized telecom AI silicon	Strategic bottleneck
Cloud dependence	Dominant	Declining at inference layer	Training stays centralized	Hybrid continuum
Model security	Emerging requirements	Attestation/tool controls	Critical network function	Mandatory for autonomous operation

The most probable five-year pathway is therefore neither the disappearance of cloud computing nor a universal migration of foundation models into antennas. 2026–2027 is the architecture-definition stage: standards bodies formalize compute-network coordination, distributed inference, semantic awareness, AI-agent terminology, NTN intelligence, AI-workload observability and security. 2027–2028 is likely to be the orchestration stage, during which operators and industrial networks begin treating compute capacity as routable infrastructure and deploy increasingly specialized inference accelerators near RAN sites and metropolitan edges. 2028–2029 should see greater adoption of digital network twins as pre-execution environments in which autonomous policies are tested before deployment, alongside bounded AI-RAN control functions and early goal-oriented communications. 2029–2030 is the probable convergence point: IMT-2030 candidate technologies, Chinese communication-sensing-computing-intelligence targets, European AI-native 6G programmes and new semiconductor architectures reach enough maturity for end-to-end demonstrations in which sensing data trigger distributed inference, autonomous agents allocate communications and computation, and physical machines respond. By 2031, the strategic distinction will no longer be between “telecom” and “cloud” infrastructure. A high-end network will increasingly behave as a geographically distributed computer whose processors happen to be connected by radio, fibre and satellite links and whose control logic is partly autonomous. The winners will not necessarily be the countries or firms possessing the single largest AI model; they will be those capable of coordinating radio spectrum, edge accelerators, memory, optical transport, cloud compute, satellite connectivity, sensing data and trusted autonomous agents as one coherent machine.

Pillar III — Surveillance, Sovereignty and the 2026–2031 Strategic Contest: Privacy, Intelligence Power and the Political Economy of Radio-Generated Data

The strategic significance of 6G sensing begins where the conventional privacy debate ends. The central issue is not simply whether future networks can detect a person, estimate a location or infer motion from radio propagation; it is whether the infrastructure that already surrounds homes, offices, factories, vehicles, airports, streets and borders acquires the institutional authority, computational capacity and commercial incentive to convert those electromagnetic observations into persistent knowledge about the physical world. Under that model, telecommunications infrastructure becomes a producer of data rather than merely a carrier of data. This creates a new sovereignty layer because control over a sensing-capable network can potentially determine who may measure an environment, who may request sensing operations, which raw observables are retained, which derived features are generated, whether identities may be correlated across time, where inference executes, and whether sensing outputs can be transferred to law-enforcement, intelligence, defence, commercial or foreign entities. The distinction between raw radio measurements and derived information is particularly important. A channel estimate may initially look like technical telemetry; after repeated observation, multi-node fusion and AI inference it can become occupancy information, trajectory data, object classification or potentially a stable behavioural signature. That is the point at which network engineering becomes political economy. ITU-T work on integrated sensing and communications already treats sensing as a formal network function involving sensing requests, sensing nodes, collaborative measurements, processing, historical information and exposure of sensing results; security mechanisms contemplated by the architecture include authentication of sensing requests and devices precisely because sensing is becoming an actionable network service rather than an incidental radio phenomenon. The trajectory therefore shifts the policy question from “who can intercept communications?” toward the much broader question “who has the right to cause infrastructure to observe an environment, and what may subsequently be inferred from those observations?” That is the core sovereignty problem of Pillar III.

The privacy frontier becomes especially consequential because the EU Artificial Intelligence Act defines biometric identification technologically neutrally rather than tying it to visible-light cameras. The Regulation describes biometric identification as automated recognition of physical, physiological or behavioural human features and expressly lists features including body shape, gait, posture and physiological characteristics; remote biometric identification is defined functionally as identification of people without their active involvement through comparison with reference biometric data, regardless of the particular technology used. This means that future RF-derived identification cannot be assumed to fall outside biometric law merely because no optical image exists. The legal analysis must examine the feature being extracted, the purpose of processing, the presence of a reference database, whether the system identifies or merely detects, and whether the processing falls within a prohibited, high-risk or otherwise regulated category. The same Regulation treats real-time remote biometric identification in publicly accessible spaces for law-enforcement purposes as particularly intrusive and permits it only under tightly defined exceptions, with judicial or independent administrative authorization, proportionality requirements, temporal, geographic and personal limits, reporting duties and deletion obligations if authorization is rejected. The Regulation also prohibits specific forms of biometric categorization used to infer particularly sensitive characteristics and prohibits untargeted scraping of facial images to build recognition databases. These provisions were drafted in a predominantly camera-centric technological era, but their language is sufficiently modality-neutral to matter for RF sensing. The harder issue lies one step below identity. A radio network may infer that three persons are present, estimate their movement and reconstruct recurring trajectories without ever naming them. That data may not automatically satisfy the legal definition of biometric identification, yet it can become privacy-invasive when correlated with access logs, devices, vehicles, schedules or historical patterns. The legal challenge between 2026 and 2031 will therefore be inference governance, not only biometric governance: the Union will increasingly need to distinguish raw radio telemetry, environmental sensing data, personal data, biometric data, derived inferences and legally sensitive automated decisions.

Data layer	Example in a 6G sensing system	Identification value	Likely governance sensitivity
Raw waveform	Received IQ/radio samples	Low in isolation	Communications/security regulation
Channel measurement	Delay, phase, Doppler, CSI	Low–medium	Potential personal/environmental data
Spatial feature	Position, velocity, occupancy	Medium	Privacy/location governance
Behavioural feature	Gait, posture, recurrent movement	Medium–high	Potential biometric relevance
Persistent signature	Stable RF-derived profile	High	Biometric/personal-data risk
Correlated identity	RF signature + access/device record	Very high	Identification regime
Predictive inference	Likely future location/activity	Very high	Profiling/automated decision concerns
Cross-site profile	Multi-location behavioural history	Extreme	Surveillance and fundamental-rights risk

The European Data Act adds a second layer that could become unexpectedly important for 6G because it governs data generated by connected products and related services, including data concerning product performance, use and environment, and requires users to be informed about what data a product is capable of generating, whether generation is continuous or real-time, where those data are stored and how they may be accessed. The Regulation explicitly recognizes data generated automatically by sensors and data reflecting the environment or interaction of connected products. Yet it also draws a crucial economic boundary: data derived through proprietary, complex algorithms — including some outputs of sensor fusion — are not automatically subject to the same mandatory access obligations as readily available source or pre-processed product data. This distinction could become one of the defining ownership battles of sensing-based 6G. If a base station or connected device generates raw channel measurements that are legally accessible to a user, but a network operator or vendor transforms those measurements through proprietary AI into occupancy maps, motion classifications, industrial process intelligence or behavioural models, the economic value may concentrate in the derived inference layer, not the raw data layer. In practical terms, the entity controlling the inference pipeline can own the most valuable part of the knowledge chain even when access rights exist for underlying data. Between 2026 and 2031, disputes over the political economy of radio-generated information are therefore likely to move away from simple “data ownership” language toward control of derivation rights, model access, semantic representations, data retention and inference portability. The broader European policy dilemma is obvious: forcing broad access to derived sensing intelligence can promote competition and innovation, but it can also increase privacy and security exposure; allowing proprietary enclosure can protect sensitive processing but create vendor lock-in and new surveillance monopolies. The Data Act already reveals the contours of this conflict by separating accessible product data from inferences generated through additional algorithmic investment.

Value-chain layer	Principal controller	Economic value by 2031	Lock-in risk
Spectrum access	State/regulator/operator	Very high	High
RF hardware	Vendor/operator	High	Medium-high
Raw sensing telemetry	Operator/device owner	Medium	Medium
Edge compute	Operator/cloud/industrial owner	Very high	High
Sensing model	AI/vendor ecosystem	Very high	Very high
Derived spatial twin	Platform/operator	Extreme	Very high
Identity correlation	Government/platform/enterprise	Extreme	Extreme
Historical behaviour graph	Data controller	Extreme	Extreme
Cross-domain fusion	Intelligence/industrial platform	Strategic	Extreme

The intelligence exploitation potential follows directly from this architecture because radio sensing creates information that is orthogonal to conventional SIGINT. Traditional communications intelligence concentrates on emissions, transmitters, protocols, metadata and content associated with communicating entities. Integrated sensing adds the possibility of extracting information about non-transmitting objects and persons from the effect they have on someone else’s signals. This changes the operational logic. In a building containing ordinary wireless infrastructure, a target may not need to carry an active handset to influence the radio environment; movement may still alter multipath, Doppler and channel state. That does not imply universal identity extraction or reliable surveillance through arbitrary structures, but it does create an additional observation modality that can be fused with access control, imagery, acoustic sensing, device metadata or human intelligence. From an intelligence perspective the key metric is therefore not whether RF sensing alone achieves certainty; it is whether the new modality materially increases posterior confidence when fused with other evidence. If visual surveillance estimates that a person entered an area but loses line of sight, radio sensing may contribute continuity; if a device identifier is ambiguous, movement correlation may strengthen or weaken attribution; if a protected facility disables cameras for privacy reasons, sensing-capable infrastructure may still create occupancy information. The most strategically valuable outputs may be mundane rather than biometric: room occupancy, perimeter movement, equipment relocation, vehicle direction, crowd-density change, drone presence, mechanical activity or patterns of life. These are precisely the types of weak signals that become powerful when accumulated. In Bayesian intelligence terms, the danger is compound observability. Each individual sensing channel may have limited confidence, but repeated multi-modal observations can drive identification or behavioural inference probabilities upward. That is why future privacy policy cannot focus exclusively on one sensor modality. A system that never “recognizes a face” may nevertheless produce a highly identifying behaviour graph after six months of cross-domain correlation.

The counter-surveillance problem is more difficult than simply disabling Wi-Fi because integrated sensing turns ambient infrastructure into the measurement system. Effective protection therefore has to operate across protocol, architecture, physics and governance. At protocol level, raw or high-granularity channel feedback should be minimized, authenticated and exposed only under explicit authorization; sensing requests should be attributable to authenticated entities and subject to purpose limitation, just as privileged API calls are today. At architectural level, sensing data should be processed as close to origin as possible when full raw measurements are unnecessary, with only derived safety or occupancy indicators exported outside the local domain. At the data layer, retention periods should be aggressively bounded and cross-context correlation should require additional authorization because a transient location estimate is fundamentally different from a persistent behavioural profile. At the RF level, privacy-enhancing waveform design, randomized reference signals, resolution limitation, deliberate feature suppression and trusted sensing modes may become future research directions, although none should be presented as universal countermeasures because physical propagation cannot simply be “encrypted away.” At the system level, the most important control may be a sensing transparency plane: individuals, organizations or regulators should be able to determine when infrastructure is operating in enhanced sensing mode, which sensing categories are active, who requested them and whether derived results are retained. This would be analogous to network observability for privacy itself. Counter-surveillance also creates an obvious security trade-off. Strong RF privacy techniques may impair legitimate industrial safety, localization, drone detection or emergency response; conversely, unrestricted sensing may create invisible mass observation. The strategic problem is therefore a controlled capability problem, not a binary enable/disable question. By 2031, high-trust 6G implementations are likely to differentiate themselves through sensing authorization architectures, while low-trust systems may be judged not by whether they communicate securely but by whether their physical-world inference functions can be independently audited.

Military applications create the sharpest form of dual-use ambiguity. The U.S. Department of Defense Electromagnetic Spectrum Superiority Strategy states that freedom of action in the electromagnetic spectrum is a prerequisite for operations across domains and integrates spectrum management and electronic warfare into unified Electromagnetic Spectrum Operations. Although that strategy predates commercial 6G standardization, its logic becomes more relevant as communications, sensing, positioning and autonomous systems converge. A future 6G-derived military architecture could use common radio infrastructure for resilient communications, passive or cooperative detection, localisation, unmanned-system coordination, distributed sensor fusion and spectrum awareness. Such systems may be especially useful in dense urban, industrial or littoral environments where deploying dedicated radar at every node is impractical. The strategic advantage comes from sensor density and dual-use ambiguity: a radio node deployed ostensibly for connectivity may also contribute measurements to a wider situational-awareness picture. This could complicate targeting law, escalation management and infrastructure protection because adversaries may view telecommunications nodes as militarily relevant if they materially contribute to command, sensing or autonomous targeting chains. Conversely, commercial infrastructure may provide useful redundancy during conflict, allowing degraded military networks to exploit terrestrial or non-terrestrial civilian connectivity. The risk is an erosion of the distinction between civilian communications infrastructure and military sensing infrastructure. China’s official 2026 programme explicitly links 6G development with satellite internet, wireless sensing, low-altitude economy and embodied intelligence, while its wider communication strategy describes future networks as deeply integrating communications, sensing, compute, intelligence and security. That does not by itself prove military integration, but it does create a technologically fungible base from which military applications can be developed. The resulting 2031 strategic environment is one in which spectrum dominance, sensing density, AI processing and infrastructure resilience increasingly interact as one military-technical system.

Military/dual-use function	6G-derived capability	Operational benefit	Strategic risk
Distributed detection	Multi-node RF sensing	Dense coverage	Civilian infrastructure becomes target-relevant
Drone awareness	Doppler/trajectory sensing	Low-altitude monitoring	False positives/adversarial spoofing
Cooperative localization	Multi-node timing/angle fusion	GNSS-denied support	Dependency on infrastructure integrity
Spectrum awareness	Dynamic occupancy mapping	EMS superiority	Intelligence collection concerns
Autonomous coordination	Low-latency AI-network control	Swarm/robotic operation	Escalation speed
Resilient C2	Terrestrial + NTN integration	Continuity under attack	Cross-domain dependency
Infrastructure defence	RF anomaly detection	Perimeter monitoring	Persistent surveillance
Digital battlefield twin	Sensor-fused spatial model	Faster decision loops	Model poisoning/misattribution

Spectrum sovereignty is the next strategic battleground because sensing-capable 6G increases the economic and military value of spectrum beyond connectivity. The United States’ National Spectrum Strategy treats spectrum access as foundational to economic growth, public safety, federal missions and global technological leadership and identifies 2,786 MHz across five groups of bands for in-depth study for potential repurposing or expanded sharing. The implementation plan includes the Lower 3.1–3.45 GHz band and 7.125–8.4 GHz, with study timelines extending into October 2026, alongside other bands. NTIA also explicitly states that spectrum management increasingly requires granular information in the time, waveform and geographic domains and calls for AI/ML-enabled dynamic spectrum sharing. The logic matters for 6G sensing: spectrum policy can no longer be modeled exclusively as allocating bandwidth to services. Sensing quality depends on bandwidth, waveform availability, coexistence constraints, antenna geometry, incumbent emissions and the ability to coordinate observations across nodes. A state that lacks access to suitable harmonized spectrum may still deploy 6G communications, but it could be structurally disadvantaged in high-resolution sensing or integrated communications-radar applications. Conversely, dynamic sharing technologies may allow countries to extract greater sensing and communications value from spectrum already occupied by defence, satellite, scientific or commercial systems. The American model therefore frames spectrum as a national strategic resource managed through data, sharing technology and coexistence engineering, not merely through auctions. This approach also strengthens international bargaining power because domestic band preferences influence positions at future World Radiocommunication Conferences and consequently the global equipment ecosystem.

China has moved even more explicitly from spectrum planning to experimental 6G allocation. On 8 May 2026, the Ministry of Industry and Information Technology authorized 6 GHz spectrum for technical 6G trials by the IMT-2030 promotion group in selected regions, specifically to support technology research, standards work and verification against ITU-defined scenarios and performance indicators. On 3 June 2026, MIIT launched a ministry–province collaborative 6G pilot programme intended by 2029 to generate domestically innovative technical solutions, new application scenarios and new terminal products while coordinating central and local industrial resources. The programme explicitly identifies integration of communications with AI, satellite internet and wireless sensing, as well as development of base stations, core networks, transport systems, chips, operating systems and commercial-space capabilities. A separate 2026–2028 MIIT programme requires at least 75% coverage of a one-millisecond metropolitan computing circle by 2028, more than 30 high-value AI-communications application scenarios, and by 2030 a major improvement in integrated communication-sensing-compute-intelligence capability. These targets reveal a sovereignty model significantly broader than “China wants 6G leadership.” The objective is vertical integration across spectrum + radio + compute + AI + satellites + devices + applications + regional industrial policy. Such integration can reduce dependence on foreign control points and creates a large domestic experimental environment in which sensing-capable applications can be tested at scale. The political consequence is that standards competition and industrial policy become inseparable. A technical architecture that wins deployment inside a very large national market can influence device economics, patents, vendor ecosystems and ultimately international standards.

Strategic layer	United States	European Union	China
Spectrum model	Dynamic sharing + federal/commercial coexistence	Harmonization + regulatory governance	State-coordinated experimental allocation
6G research emphasis	AI-native, sensing, resilience, security	Sovereignty, AI-native networks, vertical industry	Integrated communication-sensing-compute-intelligence
Governance strength	Security/standards ecosystem	Privacy/fundamental-rights regime	Centralized industrial coordination
Key advantage	Compute/software/standards ecosystem	Regulatory power + industrial research base	Scale + integrated state-industrial execution
Key vulnerability	Fragmented infrastructure ownership	Fragmented national markets/vendors	External semiconductor dependencies
Likely 2031 differentiator	Trusted adaptive infrastructure	Regulated high-trust 6G	High-scale integrated sensing ecosystem

The European Union occupies a unique position because its potential competitive advantage lies less in having the largest domestic telecom equipment market or the deepest hyperscale AI ecosystem than in defining the governance architecture under which sensing infrastructure can be considered trustworthy. The EU’s €900 million public contribution to the Smart Networks and Services Joint Undertaking for 2021–2027 is intended to be matched by private investment, creating a programme of at least €1.8 billion, while the Commission announced a further €116 million for twenty projects in March 2026 and additional planned funding through 2027. Europe’s challenge is that technological sovereignty is distributed among multiple national ecosystems rather than concentrated in a single federal-industrial structure. Germany possesses major industrial automation, semiconductor-equipment and telecom research capacity; France combines telecom, defence, space and sovereign-cloud ambitions; Italy has relevant telecommunications, aerospace, defence electronics and semiconductor assets; the Nordic states hold strong network-equipment and wireless-research positions; the Netherlands has strategic semiconductor-equipment capabilities. This diversity can become an advantage if coordinated, but fragmentation increases integration cost and slows scale. Europe’s potential strategic differentiator is therefore regulated sensing: privacy-preserving ISAC architectures, transparent sensing authorization, secure edge processing, interoperable data spaces, trusted AI inference and auditable network agents. If Europe can turn legal constraints into design requirements early enough, it can attempt to export a “trusted 6G” model analogous to previous regulatory influence in data protection. If it fails, it risks becoming a regulatory jurisdiction sitting on top of hardware, cloud, models or chip supply chains controlled elsewhere. The sovereignty contest is therefore not between regulation and innovation; it is whether Europe can encode its governance preferences into network architecture before global technical dependencies harden.

Infrastructure dependency is accordingly multidimensional. A state may own spectrum but depend on foreign base-station vendors; own the radio hardware but depend on foreign advanced semiconductors; own the chips but depend on foreign electronic-design automation or packaging; own the compute but depend on a foreign cloud orchestration layer; own the cloud but depend on foreign foundation models; own the models but depend on satellite infrastructure; own all those elements but rely on international standards and patent pools controlled elsewhere. 6G sovereignty should therefore be measured as a dependency graph rather than as the percentage of domestically manufactured base stations. Sensing raises the stakes because dependency now affects not only confidentiality and availability but also epistemic control — who determines what the network can perceive and how those observations are represented. Proprietary sensing models could encode undocumented biases; foreign cloud infrastructure could receive derived environmental data; closed radio firmware could expose measurements unavailable to the operator; external model updates could change detection behaviour without visible hardware modifications. This creates a strategic requirement for verifiable interfaces across the stack. Governments will likely push for trusted model provenance, secure boot, attested sensing functions, interoperable data formats and domestic control of high-value inference datasets. By 2031, procurement rules for critical 6G infrastructure may therefore resemble a fusion of telecom-security requirements, AI governance, cloud sovereignty and defence supply-chain controls. The key question will not simply be “is this vendor trusted?” but “which layer of observation, inference, storage or actuation does this supplier control?”

The emerging political economy of radio-generated data may ultimately become as important as the spectrum itself. In earlier telecom generations, the dominant monetizable assets were subscriptions, spectrum rights, traffic, location data and application ecosystems. In sensing-capable 6G, physical environments become continuously measurable economic objects. A factory’s radio infrastructure could generate operational maps; a logistics network could infer asset flows; a retail environment could derive occupancy patterns; a transport system could produce mobility intelligence; a building could generate high-frequency occupancy and activity models without deploying separate camera systems. The raw sensing measurement may have little standalone value, but aggregated historical data can become a proprietary dataset for training predictive models. This creates data-network effects: the operator or platform with the largest sensing footprint may accumulate the best environmental models, which improve detection accuracy, which attracts more deployments, which generate more training data. The cycle can create a new form of platform concentration. Because many derived outputs are not physically “collected” in the traditional sense but algorithmically inferred, existing contractual and regulatory assumptions about ownership may be contested. The political economy therefore shifts from extraction of user-generated content toward extraction of environment-generated information. Whoever controls the sensing model determines which latent characteristics are transformed into economically useful data. That control may become a significant source of bargaining power between telecom operators, hyperscalers, industrial companies, cities and governments.

Emerging 6G data market	Buyer/user	Commercial value	Governance risk
Occupancy intelligence	Buildings/retail	Medium	Persistent monitoring
Industrial motion map	Factory operator	High	Trade-secret exposure
Logistics trajectory model	Port/supply-chain operator	Very high	Strategic intelligence
Mobility density	Cities/operators	High	Re-identification
RF digital twin	Industrial/defence user	Extreme	Infrastructure mapping
Behavioural signature	Security/platform	Extreme	Biometric inference
Spectrum occupancy intelligence	Operator/government	Very high	Military sensitivity
Low-altitude object tracks	Aviation/security	Very high	Dual-use surveillance
Cross-site pattern-of-life	Intelligence/security	Strategic	Mass-surveillance potential

A structured Analysis of Competing Hypotheses produces six useful 2031 outcomes. H₁ — Regulated Industrial Sensing assumes ISAC scales primarily in factories, transport, robotics and private networks under strong privacy controls. H₂ — Commercial Sensing Platformization assumes operators and hyperscalers create broad sensing APIs and derived-data markets. H₃ — Security-State Expansion assumes public-security and intelligence applications become a major driver of deployments. H₄ — Sovereign Technological Blocs assumes the United States, Europe and China build materially different sensing, AI and data-governance ecosystems. H₅ — Privacy Backlash and Constraint assumes litigation, regulation and public opposition slow high-resolution human sensing. H₆ — Dual-Use Fusion assumes civilian 6G infrastructure becomes increasingly integrated into defence, emergency and national-security functions. An August 2026 Bayesian baseline places the strongest evidence behind H₁ and H₄ because industrial ISAC and geopolitical technology sovereignty are already embedded in formal programmes, while H₂ remains technically credible but institutionally uncertain and H₃ remains dependent on national law and classified or non-public capability not available under this evidence framework. A 300,000-trial exploratory Monte Carlo model using privacy regulation, spectrum availability, sensing accuracy, AI-edge maturity, geopolitical fragmentation, defence demand, public acceptance, semiconductor access and operator monetization produces an illustrative 2031 distribution of approximately 24% H₁, 18% H₂, 14% H₃, 25% H₄, 8% H₅ and 11% H₆. These values are analytical scenario outputs, not empirical probabilities supplied by the cited institutions. Sensitivity testing shows that H₄ becomes dominant when semiconductor constraints, spectrum divergence and national-security regulation correlate; H₂ becomes dominant only when standardized sensing APIs combine with weak privacy friction and strong operator monetization; H₅ rises sharply after major misuse scandals or court restrictions; H₆ accelerates after serious geopolitical conflict or sustained drone/spectrum threats.

The five-year trajectory therefore points toward a fragmented sensing order rather than one global 6G model. During 2026–2027, states will focus on spectrum studies, technical requirements, sensing-security architecture, AI governance and national test programmes. During 2027–2028, the policy problem shifts toward authorization: who can activate sensing, what data categories can be retained, whether radio-derived human features count as biometric data, and how national-security exceptions are defined. During 2028–2029, commercial and industrial platforms will begin exposing more structured sensing results, increasing the economic value of inference models and triggering competition over data portability and proprietary sensor-fusion outputs. 2029–2030 is likely to mark sharper geopolitical divergence as early IMT-2030 implementations intersect with semiconductor controls, trusted-vendor policies, satellite integration and sovereign cloud initiatives. By 2031, the decisive strategic asset will not merely be a 6G radio network but a trusted or controlled physical-world information system capable of sensing, interpreting and acting upon its environment. Europe is likely to emphasize legitimacy, privacy and auditable trust; the United States dynamic spectrum sharing, AI, security and private-sector innovation; China state-coordinated scale, industrial integration and communications-sensing-compute-intelligence convergence. None of these models is predetermined to dominate globally. The key competitive variable will be whether each bloc can transform its normative preferences into deployable technology rather than depending on another bloc’s hardware, software, models or data architecture.

2026–2031 Strategic Contest Matrix

Vector	2026	2027	2028	2029	2030–2031
Privacy law	AI/GDPR rules applied to emerging sensing	RF-specific interpretations	Case-law and guidance	Architectural compliance	Jurisdictional divergence
Biometric inference	Research/early systems	Regulatory classification	Controlled deployments	Cross-modal fusion	Selective operational maturity
Sensing APIs	Early architecture	Prototype exposure	Industrial APIs	Commercial platforms	Differentiated ecosystems
Spectrum sovereignty	Band studies/trials	Harmonization competition	Pre-commercial planning	Allocation decisions	Strategic divergence
Intelligence exploitation	Experimental/adjacent capability	Multi-modal fusion growth	Operational niche systems	Persistent sensing in selected domains	High-value intelligence layer
Counter-surveillance	Protocol research	Authorization controls	Privacy-enhancing architectures	Auditing standards	Trusted-sensing certifications
Military integration	Dual-use experimentation	C-UAS/EMS integration	Cooperative sensing	Multi-domain fusion	Selective civil-military convergence
Data markets	Raw IoT/data services	Derived sensing services	Platform consolidation	RF digital-twin markets	Environment-data economy
Sovereignty policy	Vendor trust	AI/cloud dependency scrutiny	Chip/model provenance	Full-stack procurement	Bloc-specific 6G stacks

---

Copyright of debuglies.com – Even partial reproduction of the contents is not permitted without prior authorization – Reproduction reserved