Next-Generation Networks

A “next-generation network” in September 2026 means four things at once, and they run on different clocks. Established The first is 5G–Advanced, which is deployed commercial equipment running 3GPP Release 18 and 19 features on existing sites. The second is 6G, which is a study item: the ITU-R framework recommendation M.2160 was agreed in November 2023, the technical performance requirements stage followed in 2026, and 3GPP began its Release 20 6G studies in 2025 with normative work expected in Release 21. The third is the non-terrestrial layer — satellites talking to unmodified handsets — which has crossed from demonstration into commercial service faster than any part of the terrestrial roadmap. The fourth is not radio at all: spectrum awards, tower siting, fibre backhaul, grid connections and municipal permitting, which is where the schedule actually breaks.

Frontier The organising claim of this brief is that the binding constraint on the next generation is coverage economics in the upper mid-band, not radio physics. Every capability in the IMT-2030 framework is demonstrable in a laboratory today; none of them is demonstrably affordable on the tower grid that exists. The candidate 6G band — roughly 7 GHz to 15 GHz, the range the industry calls FR3 — sits above the 3.5 GHz bands that carry most 5G capacity, and the uplink budget there is worse. Whether the same towers can cover the same area decides whether 6G is an equipment refresh or a second national build-out, and nothing published independently has settled it.

Established Three further claims this brief defends from the measured record. First, 6G will be an upper-mid-band system rather than a terahertz one: the terahertz demonstrations are real, and they are point-to-point links over metres to hundreds of metres, limited by power-amplifier output and blockage. Second, direct-to-device satellite service is a coverage layer, not a capacity layer — its delivered bits per square kilometre are orders of magnitude below a terrestrial cell, which is arithmetic about beam footprints and does not improve with better satellites. Third, open RAN has been achieved contractually more often than operationally.

Established Scope boundaries. Municipal sensing and the procurement pathology around it belong to Smart Cities; cold-climate siting and the cost of building north of the tree line belong to Arctic Engineering; links measured in astronomical units belong to Deep Space Communications. This brief covers the terrestrial and near-Earth network: radio access, spectrum, transport, timing, sensing, and the institutions that allocate them.

Established A note on sourcing. This brief was commissioned in September 2026 from the Institute’s research base. Reading-list entries without links are cited from the bibliographic record rather than re-fetched, and claims are dated no later than early 2026 unless carried by a linked source.

Established The standards clock is public, slow, and routinely misquoted. ITU-R Recommendation M.2160, agreed in November 2023, defines IMT-2030 through six usage scenarios: immersive communication, massive communication, hyper-reliable low-latency communication, ubiquitous connectivity, artificial intelligence and communication, and integrated sensing and communication. It attaches capability ranges rather than single targets — peak rate in the tens to low hundreds of gigabits per second, user-experienced rate in the hundreds of megabits per second, latency in the 0.1 ms to 1 ms band, connection density up to the order of 10 million devices per square kilometre, positioning accuracy in the 1 cm to 10 cm range. Ranges, not promises: the framework says what a candidate may be assessed against, not what a network will deliver in a suburb.

Frontier The 2026 step was requirements, not technology. The ITU published the IMT-2030 technical performance requirements in March 2026, starting a submission-and-evaluation cycle that historically runs three to four years before specifications freeze. 3GPP opened 6G study items in Release 20 in 2025, with normative work expected in Release 21 and a first freeze in the 2028–2029 window on the schedules circulating in 2026. Anyone quoting “6G in 2030” is quoting the earliest plausible date of first commercial service in one or two lead markets, not coverage.

Established Spectrum is the decision that has already been half-made. WRC-23 identified upper 6 GHz spectrum (6425–7125 MHz) for mobile in parts of the world, with regional extensions at 3.3–3.8 GHz and 4.8–4.99 GHz, and put the upper mid-band on the WRC-27 agenda — 7.125–8.4 GHz and 14.8–15.35 GHz being the bands the 6G case rests on. The consequence is structural: the band plan will be decided by a treaty conference in 2027, before the radio specification is written, and the incumbents there — fixed satellite services, radio astronomy, government and military users — are the parties whose displacement sets the timetable.

Frontier The upper mid-band is where the honest disagreement is. Vendors argue that very large arrays — hundreds of elements, since element spacing scales with wavelength — recover the extra path loss and permit co-siting on the existing grid. The counter-argument is that array gain helps the downlink far more than the uplink, because the handset cannot grow an array or its transmit power, and modern cells are uplink-limited at the edge. Both positions are argued from simulations and vendor trials. Neither side has published an independent co-sited measurement, which is why this brief treats the question as open.

Established The terahertz reality check is a link-budget problem with a power-amplifier cause. Laboratory links above 100 GHz have carried hundreds of gigabits per second, and the D-band (110–170 GHz) is genuinely useful for short fixed backhaul hops. The obstacle is the transmitter: solid-state amplifier output falls steeply with frequency, so a 300 GHz front end delivers milliwatts where a sub-6 GHz one delivers watts, and the deficit must be repaid with antenna gain, which means pencil beams, which means blockage by a hand or a passing van. Add molecular absorption peaks near 183 GHz, 325 GHz and 380 GHz, and the plausible role of terahertz is fixed bridges and in-room links, not coverage.

Established Direct-to-device moved from stunt to service in under three years. AST SpaceMobile’s BlueWalker 3 test satellite made the first direct voice call to an unmodified smartphone in April 2023 and a broadband-class connection later that year; SpaceX’s Direct to Cell payloads carried the first satellite-to-handset text messages in January 2024 and entered commercial messaging service with a US mobile operator in 2025. The constellation is the enabling asset: several thousand satellites, tracked publicly launch by launch, with the operator’s own semi-annual filings reporting network health, failure rates and tens of thousands of collision-avoidance manoeuvres per six-month period. The optical inter-satellite links carrying traffic between them descend from the space laser-communication demonstrations of the preceding decade.

Established The capacity arithmetic is the part the marketing omits. A terrestrial cell serves a footprint of single-digit square kilometres in suburbia and a fraction of that downtown. A direct-to-device beam covers thousands of square kilometres and shares one channel across all of it. Even granting generous spectral efficiency, delivered capacity per square kilometre differs by three to four orders of magnitude, and the gap is set by geometry and licensed reuse, not by satellite design. The honest description is a ubiquitous low-rate layer: messaging, emergency traffic, telemetry and narrowband machine data outside terrestrial coverage.

Frontier Open RAN is a real architecture with a thin at-scale evidence base. The O-RAN ALLIANCE specifications, principally the open fronthaul interface between radio unit and distributed unit, exist and are implemented. Greenfield operators — Rakuten Mobile in Japan from 2020, DISH in the United States from 2023 — built national networks on them, and a multi-billion-dollar US brownfield commitment followed in December 2023. What has not been demonstrated at national scale is the promised outcome: a live macro network mixing radio units and baseband from genuinely different suppliers, with published cost and energy comparisons against the integrated alternative. Most deployments described as open RAN are single-vendor stacks that expose standard interfaces.

Established The 5G business case did not land where it was sold, and did land somewhere else. Consumer average revenue per user did not rise materially in major markets after launch, and operator revenue growth in developed markets has run at low single digits against hundreds of billions of dollars of 5G capital expenditure. The application that worked was fixed wireless access: millions of households replaced fixed broadband with a 5G modem, using capacity that was otherwise idle. That is the one large-scale commercial validation of mid-band 5G, and it was not in the original use-case deck.

Frontier Integrated sensing and communication is standardised earlier than it is understood. 3GPP extended its channel-model work to sensing in Release 19 and carried ISAC into the Release 20 studies. The measured record is short-range and bistatic: presence and motion detection, coarse localisation of people and drones, decimetre-class range accuracy when bandwidth is generous. Monostatic sensing — a base station listening to its own echo — needs transmit-to-receive isolation of roughly 100 dB to avoid deafening the receiver, and that hardware problem separates the demonstrations from a deployable feature.

Established Positioning improved more than most operators exploited. 3GPP added positioning reference signals in Release 16, refined them in Release 17 for industrial indoor use, and added carrier-phase techniques in Release 18. Field results in commercial networks typically land in the one-to-several-metre band rather than the sub-metre target, because accuracy depends on how many base stations a device can hear and on line-of-sight geometry that urban deployments do not provide. Centimetre positioning outdoors still comes from satellite navigation with real-time kinematic corrections; the network’s contribution is mostly indoors and in tunnels.

Established Timing is a hidden dependency, and it is currently a security problem. Radio networks need alignment in the tens of nanoseconds for time-division duplex and positioning, and they get it overwhelmingly from satellite navigation receivers at the cell site. Jamming and spoofing rose sharply from 2023 in several regions, to the point of standing civil-aviation warnings. The physical-layer alternative exists — optical clocks now reach systematic uncertainties around the nineteenth decimal place, and fibre transfer can distribute that stability — but a national holdover architecture independent of satellites is a programme nobody has funded at scale.

Frontier Resilience is now measured in cable incidents rather than availability targets. Submarine cable damage in the Baltic, the Red Sea and the Taiwan Strait between 2023 and 2025 removed large fractions of regional capacity for days to weeks, and in several cases cause and intent remain contested. The engineering lesson is unambiguous even where attribution is not: route diversity, spare capacity and repair-ship availability determine outcome, and repair-ship capacity is a small, ageing, globally shared fleet.

Frontier Can the upper mid-band be covered from the existing grid? The cost of 6G turns on this, and it decomposes into three measurable parts: how much array gain is realisable on a commercially mountable panel at 7–8 GHz; how much of that gain reaches the uplink from a handset limited to about 23 dBm; and how much of the residual gap can be closed by uplink repetition, which trades throughput for range.

Frontier Is there a paying application above mobile broadband? Every generation since 3G has been sold on an application other than faster browsing, and in each case revenue arrived from ordinary data growth instead. The 6G candidates are sensing-as-a-service, network exposure through application programming interfaces, deterministic industrial connectivity and non-terrestrial coverage. The interfaces exist and the joint ventures selling them formed in 2024; reported revenue is small relative to the capital at stake.

Speculative Will sensing be accurate enough to be useful and coarse enough to be lawful? The interesting regime is one where a network detects a fall, a drone or an intruder without identifying a person. Whether that regime is wide or empty is an empirical question about resolution, unanswered by measurement in a residential environment.

Established Power-amplifier efficiency above 7 GHz is the first hard wall. Base-station amplifiers lose efficiency as frequency rises and as the signal’s peak-to-average ratio grows; wider channels worsen both. Gallium nitride devices carry the mid-band today, and extending that device technology to 7–15 GHz at acceptable efficiency and cost is a semiconductor programme, not a standards item.

Established The uplink asymmetry is structural. A base station may transmit hundreds of watts of effective radiated power from a large array; a handset transmits about 200 mW from one or two antennas under a thermal budget set by a phone that must stay comfortable to hold. Every generation has widened this gap and patched it with uplink repetition, supplementary lower-band uplinks and carrier aggregation. The patches cost spectrum or throughput.

Established Fronthaul capacity scales with antennas, not with users. Splitting the radio unit from the baseband means moving digitised radio samples across a fibre whose rate grows with bandwidth and antenna streams. Larger arrays in wider channels push open-fronthaul links towards rates needing new optics at every cell site, which converts an architecture choice into a civil-works programme.

Frontier Site power and grid connection are now gating in dense markets. A heavier array on a rooftop usually means a new power feed, a structural assessment and a lease negotiation. Where capacity is scarcest these steps take longer than equipment lead times, and they are why rollout curves flatten well before spectrum is exhausted.

Established Fibre is the precondition, and it is the item most often assumed. Dense radio means dense backhaul: a site without fibre or a high-capacity microwave path has a capacity ceiling regardless of its radio. The fibre-to-the-site share is the best single predictor of whether an operator can use new spectrum, and it varies enormously between and within countries.

Established Radio-frequency semiconductors gate the band plan. Filters, front-end modules and amplifiers for a new band take years to design, qualify and yield, in a supply base concentrated in a handful of firms. A band identified at a treaty conference in 2027 becomes a volume handset component several years later; that lag, not the specification, sets first-device timing.

Frontier Channel models above 7 GHz are a scientific dependency. Propagation, building penetration, foliage loss and blockage statistics at 7–15 GHz are measured less thoroughly than at 3.5 GHz or millimetre wave. Any 6G evaluation is only as good as those models, and they come from a small community of measurement groups still running campaigns in 2025–2026.

Established Cryptographic migration is a dependency with a deadline. Post-quantum signature and key-establishment standards were finalised in 2024, and network equipment, subscriber credentials and roaming interfaces all need migration paths. Long-lived infrastructure keys are the exposure: an element deployed in 2027 may still carry traffic in 2042.

Frontier The decisive measurement is an independent, co-sited comparison of uplink coverage at 7 GHz against 3.5 GHz on the same towers, with commercial handset transmit power and a published methodology. Everything expensive about 6G follows from the answer. If the coverage radius at equal edge throughput is within roughly 10–20% of the 3.5 GHz radius, 6G is an overlay on the existing grid and its capital cost is an equipment refresh. If it is half, the upper mid-band requires a second grid, and the business case must find a revenue source mobile broadband has never provided. The trial equipment exists; what does not exist is a published result from a party without a commercial interest in the answer. The WRC-27 agenda makes the window explicit: the measurement is most useful before the band is allocated, not after.

Frontier The second test is multi-vendor fronthaul interoperability under live traffic. The open RAN claim is specific and therefore falsifiable: radio units from supplier A working with baseband from supplier B in a commercial macro network, with published figures for energy per bit, feature parity and fault-resolution time against an integrated control site. Plugfests demonstrate interoperability in a laboratory; the missing evidence is a live network run this way for a year with numbers attached.

Frontier The third is a loaded-capacity measurement of direct-to-device service. Published demonstrations report peak rates to a single handset under ideal geometry. The number that determines the technology’s role is aggregate throughput delivered to many simultaneous users inside one beam footprint, measured over a day, with the beam’s area stated. An operator or regulator could publish this from existing commercial service; none has.

Speculative The fourth is a false-alarm study for network sensing in a residential setting. Detection probability is reported; false-alarm rate in a real street, across weather, traffic and pets, is not. Sensing features will be judged on nuisance alarms, and the regulatory question of what resolution a network may lawfully achieve cannot be argued without that measurement.

Frontier The fifth is a satellite-navigation denial exercise on a live network. Several regions have been running an involuntary version since 2023. A deliberate, instrumented version — disconnecting satellite timing from a metropolitan cluster for days and publishing what degrades, when — would price the holdover architecture that resilience arguments currently assume.

Established The array is the product. A 7–15 GHz base station is defined by how many antenna elements fit behind a panel of a given wind-loading area, how efficiently each element’s amplifier runs, and how much heat the assembly sheds without a fan. Element count rises as the square of frequency for a fixed aperture, so calibration, beam management and thermal design scale faster than the radio problem.

Frontier Energy targets must be specified per bit and per site, not per network. Vendor claims of large generation-over-generation gains are usually stated per bit at full load, which flatters a network that spends most hours lightly loaded. The targets that matter are idle-mode power, sleep-state transition times and the latency penalty those transitions impose on the first packet.

Established Satellite payloads face the same arithmetic in a harder package. A direct-to-device satellite needs a very large aperture to close the link to a handset, which means deployable arrays of tens to hundreds of square metres, driving launch mass, attitude control and solar power. Optical inter-satellite links move traffic without ground stations everywhere, at the cost of pointing precision measured in microradians.

Established Optical inter-satellite links connect this brief to deep-space practice. The laser terminals that now form the backbone of large low-orbit constellations descend from the same demonstration lineage as space laser communication generally, and pointing, acquisition and tracking is the shared hard part. Deep Space Communications covers the regime where the link budget, rather than the spectrum licence, is the whole problem.

Established Spectrum coordination now extends beyond Earth. Lunar communication and navigation architectures require their own band identifications and international coordination, negotiated through the same treaty machinery that will allocate the 6G bands. The relevance is procedural: the same scarce agenda time at a world radiocommunication conference serves both.

Frontier Data-centre networking is diverging from carrier networking. Machine-learning clusters are the fastest-growing consumer of high-speed optics, and vendors building co-packaged optical switches for them optimise for power per bit over metres rather than carrier transport. Component volumes, and therefore cost curves, are increasingly set by the data centre rather than the telephone network.

Frontier Municipal deployment is where network capability meets evaluation failure. Sensing, positioning and edge intelligence are sold to cities as outcomes. The record in Smart Cities is that such systems are bought inside bundles that destroy attribution before measurement is attempted — the most likely fate of network sensing as a municipal product unless evaluation rights are written into procurement.

Established Two institutions, with different memberships, set the outcome. The ITU-R allocates spectrum by treaty at world radiocommunication conferences roughly every four years, with national administrations voting. 3GPP writes the specifications through a consensus process that gives large vendors and operators disproportionate weight. A capability becomes real only when both tracks converge, which is why the useful unit of forecasting is the conference cycle and the release cycle, not the press release.

Established Auction authority is a live institutional variable. The United States lost its statutory authority to auction spectrum in March 2023 and regained it through legislation in 2025 that also mandated a large tranche of new auctions. Nothing about the physics changed in those two years; the pipeline of assignable spectrum did. Institutional capacity to run an award determines deployment as much as technology does.

Frontier Vendor exclusion has reshaped the supply base and the cost base. Security-driven restrictions on particular equipment suppliers forced swap-out programmes costing billions, reduced the number of bidders in affected markets, and are one of the strongest structural arguments made for open RAN. The security rationale and the competition effect are separable questions, and public debate routinely merges them.

Frontier Direct-to-device breaks the separation between mobile and satellite licensing. Using terrestrial mobile spectrum from orbit requires either a supplemental coverage authorisation or a satellite operator acquiring the rights outright. Spectrum transactions in 2025, in which a satellite operator bought terrestrial mobile spectrum, indicate the second route is live, and it changes who the incumbent is in a future auction.

Frontier Network sensing turns infrastructure into an observation system without a consent event. A mobile network already knows where devices are; a sensing network detects objects carrying no device, including people who never agreed to anything. The legal categories — communications data, personal data, surveillance — were written for a network that only tracked subscribers, and the gap is not hypothetical if ISAC ships as a base-station feature.

Established Satellite emergency messaging is a genuine and unevenly distributed safety gain. Direct-to-device messaging works where terrestrial coverage does not, which is where people get into trouble. The distributional question is whether it becomes a default capability of ordinary handsets or a premium subscription, and that is decided by commercial terms rather than physics.

Established The coverage gap is mostly an affordability gap. The population living inside mobile coverage but not using mobile internet is several times larger than the population outside coverage altogether. Any next-generation programme justified on inclusion grounds should be assessed against handset and data prices first, because that is where the numbers are.

Established Connectivity has become the substrate other infrastructure assumes. Payments, logistics, grid control, water systems and emergency services now fail together when the network fails, because each removed its own dedicated communications years ago on cost grounds. This is a genuine civilizational dependency, created by rational local decisions with nobody accounting for the correlated risk.

Frontier A satellite layer changes the geography of failure rather than removing it. Satellite backup breaks the correlation with terrestrial cable cuts and local power loss, and creates a new correlation with a small number of constellations, ground segments and orbital regimes. Whether that is a net gain depends on how many independent operators exist, which is a competition-policy fact rather than an engineering one.

Frontier Timing dependence is the underrated systemic risk. A week of denied satellite navigation over a large region would degrade radio networks, financial timestamping and grid protection simultaneously. The mitigation is unglamorous, cheap relative to a spectrum auction, and unfunded almost everywhere.

These horizons track the network generation actually being deployed, not the earliest laboratory demonstration of any feature within it.

• 10 yr: Frontier 6G specifications frozen in the late 2020s, first commercial launches around 2030 in lead markets, meaningful coverage in the early 2030s. Upper-mid-band spectrum awarded following WRC-27. Direct-to-device becomes a standard handset capability for messaging and low-rate data. Open RAN either produces its first audited multi-vendor macro network or quietly becomes an interface convention inside single-vendor stacks.
• 25 yr: Speculative Sensing as a routine network service where a lawful resolution envelope has been defined, with municipal and industrial applications ahead of consumer ones. Satellite and terrestrial access wholesale-integrated so the user never learns which layer carried the packet. Terahertz confined to fixed links, in-room connectivity and instrumentation.
• 50 yr: Speculative Spectrum management substantially automated and dynamic, with database-coordinated sharing displacing exclusive licensing for most bands. Timing distributed from a resilient terrestrial architecture with satellite navigation as one input among several rather than the single point of failure.
• 100 / 250+ yr: Handwave Claims about planetary-scale integrated communication and sensing fabrics, or networks spanning cislunar space as a routine extension of terrestrial service, extrapolate an institutional process whose four-year conference cycle has not obviously changed in fifty years. The physics permits them; nothing in the record supports a schedule.

• Depends on Nothing on the Institute’s map blocks this topic in the strict sense — the dependencies are industrial and institutional rather than scientific. The nearest map dependencies are evidentiary: the procurement and evaluation pathology documented in Smart Cities, which predicts how network sensing will be bought and why its benefits will go unmeasured, and the cold-region cost structure in Arctic Engineering, which sets the floor price of terrestrial coverage in exactly the places the satellite layer is supposed to serve.
• Requires (not on this map) A supply base that can yield power amplifiers, filters and front-end modules at 7–15 GHz with efficiency and cost comparable to today’s mid-band parts, in handsets as well as base stations. Civil capacity to deliver new power feeds, structural assessments and lease amendments at tens of thousands of existing sites, the step that sets rollout pace in dense markets. A spectrum award process that actually relocates or shares with the fixed-satellite, scientific and government incumbents of the upper mid-band, since identification at a treaty conference is not assignment. A service someone pays for that is not faster browsing, because the revenue history of the last three generations says broadband alone will not fund a second grid. An independent, repeatable field measurement of uplink coverage at the candidate band against 3.5 GHz on the same towers, which no disinterested party has published. And a live multi-vendor fronthaul deployment carrying commercial traffic long enough to produce comparable cost, energy and fault-resolution figures.
• Enables Ubiquitous low-rate connectivity for emergency messaging and machine data, including at sea and in the high latitudes; deterministic industrial connectivity as a substitute for cabling inside factories and ports; indoor positioning good enough for emergency response and asset tracking; and passive environmental sensing from infrastructure already installed for another purpose.
• Adjacent Deep Space Communications shares the optical terminal lineage and the link-budget discipline; Smart Cities is where sensing and positioning become municipal products; Arctic Engineering governs what a remote site costs to build and keep standing.

Handwave “6G means terahertz.” The most durable claim in consumer coverage of the topic, and it inverts the engineering consensus. The band plan being negotiated for WRC-27 is 7–15 GHz. Terahertz work continues for fixed links, imaging and instrumentation, where propagation is bounded and the transmitter can be plugged into a wall. The claim survives because a bigger number reads as a bigger generation.

Handwave “Satellite direct-to-device will replace rural networks.” The arithmetic in section 2 answers it: a beam covering thousands of square kilometres cannot deliver the per-area capacity of a terrestrial cell, and no satellite design changes that by three orders of magnitude. What direct-to-device does replace is the assumption that a coverage gap must be filled with a tower, which matters for emergency obligations and for the economics of the last few percent of population coverage.

Frontier “Open RAN cuts total cost of ownership by a third.” Figures in this range circulate from vendors and consultancies with an interest in the transition. The evidence base is greenfield operators whose costs differ from an incumbent’s for many reasons besides architecture, and the strongest public case for open RAN is supplier diversity and security-driven vendor replacement, not a measured saving. The honest statement is that no audited at-scale comparison has been published.

Established “5G was a failure.” Half true, and the false half matters. Consumer revenue did not respond, and most advertised applications did not materialise. Fixed wireless access did materialise, at the scale of millions of households, and is the clearest commercial validation of mid-band capacity. The lesson for 6G is not that generations do not pay, but that the paying application is not the one in the launch deck.

Frontier “Edge computing will move the cloud to the cell site.” Operator-hosted edge zones were built on the premise that low latency would pull workloads out of central data centres. Uptake was weak, several hyperscaler-operator edge offerings were retrenched, and the workloads that grew instead concentrated in very large facilities for reasons of power and interconnect. Edge intelligence is real for a narrow class of latency-bound and sovereignty-bound applications, and it is not where the compute is going.

Speculative “Network sensing will let operators see inside your home.” Imaging resolution scales with bandwidth and aperture, and a cellular base station has neither in the quantities required for anything resembling a picture. The realistic capability is coarse presence, motion and coarse localisation. The privacy problem is real, and it concerns non-subscribers being detected at all rather than imaging.

Handwave “AI will run the network autonomously.” Learned components inside the radio have real measured gains in narrow functions, and vendor claims of large network-level energy savings rest on trials that are not independently reported. The step that works by assertion is the leap from a feature improving a metric in one cluster to a network managing itself across thousands of sites, multiple suppliers and a regulatory obligation to explain outages.