What Will Actually Change in 6G?

Generational questions are usually answered with a capability table — peak rate, latency, connection density, one column per generation. It is the wrong instrument. Nobody's work changed when the peak rate figure went up.

The question worth asking is narrower and more uncomfortable: which of the things you currently assume will stop being true? Run 5G through that filter and the answer is smaller than the marketing, and more interesting.

---

Start with what does not change

This is the part that gets left out, and it is most of the system.

The waveform. OFDM with a cyclic prefix stays. It is not exciting and nothing has displaced it; the alternatives studied for a decade have not beaten it on the combination of complexity, MIMO compatibility and equalisation cost that made it win in the first place.

The frame structure. Slots, numerologies, scalable subcarrier spacing. The specific numbers may extend upward for higher bands, but the shape carries over.

HARQ, and the scheduler above it. The retransmission machinery and the per-slot allocation loop are generation-independent problems with settled answers.

The core. A service-based architecture with sessions and QoS flows is where 5G ended up after a genuine break from 4G. There is no comparable break queued behind it.

And the spectrum you already own. Every generation refarms the low bands, because propagation does not care what the standard is called. Sub-7 GHz will carry 6G coverage exactly as it carries 5G coverage.

If you know the 5G air interface, you will recognise almost all of a 6G one. What follows is the part you will not.

1. The network produces something that is not data

This is the largest change, and it is architectural rather than radio.

Integrated sensing and communication means the same waveform and the same infrastructure that carry traffic also return a picture of the physical environment — objects, movement, position. The physics is well understood; radar has done it for eighty years.

What is not settled is everything around it. A sensing result is an output the mobile network has never had before, and it needs somewhere to go: an interface to expose it on, a consumer authorised to receive it, a retention policy, a legal basis. The network will detect people who are not subscribers and never agreed to anything, which is a category of problem no existing 3GPP interface was designed for.

Expect the technical work to be the easy half.

2. Models become network elements

5G-Advanced already puts machine learning inside specific functions. What changes in 6G is that a model becomes a thing the network operates, with the operational baggage that implies.

If channel estimation, beam selection or CSI compression runs on a learned model, then that model has a version, a training set, a validity region, and a failure mode that is not a crash but a quiet loss of accuracy when the environment drifts away from what it was trained on. It needs monitoring, rollback and a defined behaviour when it is switched off mid-session.

Harder still: some of these models sit at both ends. A CSI compression model in the device and its decoder in the network have to match. That makes model distribution a two-sided interoperability problem across vendors — closer to a release-engineering discipline than to radio design, and genuinely new for this industry.

This is why AI-native is a useful phrase despite the wear on it. The distinction is not that models exist. It is that data pipelines, lifecycle and governance become architecture.

3. Upper mid-band re-opens site planning

The spectrum getting the most serious attention is not sub-THz. It is 7–24 GHz, the upper mid-band, sometimes called FR3.

It matters because it is the last place where a large contiguous block might come with tolerable propagation. And it is disruptive because it sits in an awkward gap: too high for the existing macro grid to cover the same way, too low to be treated as the short-range special case FR2 became.

The practical consequence is that coverage at FR3 depends on the array closing a link budget the site geometry no longer closes on its own. Massive MIMO stops being a capacity feature and becomes a coverage prerequisite. Planning assumptions calibrated at 3.5 GHz do not transfer, and neither do the site-count economics built on them.

Which is also why WRC-27 matters more than any technical milestone in the calendar. Until the bands are allocated, none of this planning work has fixed inputs.

4. Energy becomes a quantity you schedule

Energy efficiency has been a stated goal for two generations and an afterthought in both. What changes is that it becomes a constraint the network actively manages rather than a number reported afterwards.

The mechanics are already visible in 5G-Advanced: switching off components, cells and carriers on real timescales. The obstacle is that a cell cannot sleep while it is obliged to transmit always-on signals — synchronisation blocks, broadcast information — on a fixed period, because idle devices depend on finding them.

So the interesting work is in reducing what a cell must transmit when nobody is listening: longer or adaptive periodicity for the always-on signals, broadcast information delivered on demand, deeper sleep states with defined wake latency. These are not efficiency tweaks; they change what a device can assume is present, which reaches all the way back into initial access.

5. Satellite access stops being an overlay

Non-terrestrial networks exist in 5G from Release 17, as a bolt-on with its own conditions attached. The 6G intent is one system where the access happens to be in orbit.

The engineering consequence is mobility. A terrestrial cell is fixed and the device moves. A low-Earth-orbit cell moves at several kilometres per second regardless of what the device does, so handovers become scheduled events driven by ephemeris rather than reactive ones driven by measurements. Timing advance and Doppler compensation become continuous corrections rather than occasional ones.

Nothing about that is impossible. It is simply a different set of assumptions than every terrestrial mobility procedure was written under.

When any of this lands

The honest shape of the timeline is continuity, not a step. Release 18 brought AI/ML for the air interface and network energy saving. Releases 19 and 20 extend that and add sensing and positioning study work. Release 21, from around 2026 to 2028, is expected to carry the first normative 6G specifications, with commercial deployment near 2030.

Notice what that means: every one of the five changes above is being de-risked inside 5G-Advanced right now. That is where the specification work is visible, where the trials are, and where an engineer can usefully pay attention today. Waiting for a 6G launch to start reading is waiting for the summary.

And the first release will be modest, as first releases always are. Early 5G was non-standalone with a fraction of the feature set; the capabilities people now associate with the generation arrived years later. There is no reason to expect a different shape this time. The broader 6G picture — the agreed targets, the standardisation process, and which claims deserve scepticism — is worth reading alongside this.

---

The mental model

Most of the radio survives. OFDM, slots, HARQ, the scheduler, the service-based core, and the low bands you already own.

Sensing is the real architectural change, and its hard problems are interfaces, consent and retention rather than physics.

Models become operated elements with versions, drift and two-sided interoperability — a release-engineering discipline, not a radio one.

Upper mid-band re-opens planning. At 7–24 GHz the array closes the link budget the geometry no longer does.

Energy management reaches into always-on signalling, which changes what a device can assume is there.

NTN mobility is scheduled, not reactive, because the cell is what moves.

None of it arrives as a step. It matures in 5G-Advanced and gets architected properly in the first 6G release.

If you take one thing: the generation label will change on a fixed date, and none of these five will.

Further reading

• ITU-R Recommendation M.2160 — the IMT-2030 framework and capability set
• 3GPP Release 18 — AI/ML for the NR air interface, network energy saving, NTN enhancements
• 3GPP Release 20 — 6G study items, and the ISAC and positioning studies alongside them
• WRC-27 agenda items covering the upper mid-band
• Hexa-X-II and the other regional 6G flagship programmes