What pays before the freeze

A capability earns nothing until its release, its band and its device all exist

Before a capability earns anything, three things have to be true at once. Somebody has to have frozen the specification that defines it, somebody has to hold a licence for the spectrum it runs in, and somebody has to be selling a device that uses it. Different bodies control each of the three, which is why they slip independently, and each has a date on the record (Exhibit 1).

Start with the specifications, because 3GPP publishes its dates and they are the easiest to check. Release 17 froze in March 2022, Release 18 in March 2024 and Release 19 in December 2025. Release 20 is projected to freeze in March 2027. The one that matters for 6G is Release 21, the first to carry normative 6G specifications. It is planned to freeze in December 2028, with its code freeze three months later in March 2029 (part 1).

Spectrum is slower, and its dates come from a statute, an FCC order and a conference rather than from one body. The bands a carrier holds today, the C-band and 3.45 GHz, have been licensed since 2021 and 2022, so for them this dependency is already met. The upper C-band is next. Auction 115 opens for bidding on 27 April 2027, but the clearing comes after the sale. Terrestrial service may not begin until 31 December 2030 in the top 75 areas, and 1 July 2031 everywhere else. Above 4.2 GHz nothing is dated at all. An NTIA report is due in December 2026, WRC-27 meets in Shanghai from 18 October to 12 November 2027, and the statute then allows an American auction as late as July 2033 (part 2).

Devices are the dependency people forget, and the only evidence we accept for one is a published commercial launch. T-Mobile put the first commercial RedCap device in North America on sale on 17 October 2024, and AT&T declared nationwide RedCap coverage on 16 July 2025. Direct-to-device phones are in service, as the direct-to-device series records. The upper C-band is the odd one out, because its device date is a rule rather than a launch. Note 12 to TS 38.101-1 restricts band n77 in the United States to 3450 to 3550 and 3700 to 3980 MHz. That is another way of saying that no American handset is specified for 3.98 to 4.14 GHz today. The FCC’s order requires every new device sold for the band to cover it from 31 December 2030, so the handsets have four years to catch up.

Five candidates have all three today, and all five are 5G features

Score the ten on those three dependencies and they fall into three groups without much argument. Five earn now. Two more are dated, one by its spectrum and one by its devices, and the last three have nothing to build to until Release 21 freezes (Exhibit 2).

The five that earn now are all 5G features, and most of them are old ones. Fixed wireless, private networks and direct-to-device come from Releases 15, 16 and 17. Is there money in them? Fixed wireless is the one with the hardest numbers. T-Mobile reported 8.45 million 5G broadband customers at the end of 2025, Verizon 6.2 million fixed wireless subscribers at mid-2026, and the list prices run from $35 to $70 a month. Those two counts are, as far as we can find, the only hard evidence anywhere of what mid-band capacity sells for. None of the five appears on a 6G slide under its own name. They are nonetheless what a carrier earns with across the whole window the 6G plan covers.

The energy row is the nearest any of the five comes to the AI-native story that part 5 examines. Release 18 introduced network energy saving and Release 19 enhanced it, in work item RP-240170. What a carrier buys is software for radios it already owns. No new band is involved and no architecture changes, which is why we count vendor software as the device.

Two candidates are dated by spectrum or devices, and three wait for Release 21

The date a capability earns is the date its last dependency arrives, whichever one that is. For the upper C-band the laggard is the licence and the clearing; for XR it is a device and a service; and for the last three rows it is the specification itself, which does not yet exist (Exhibit 3).

Take the upper C-band first, because it is the easiest case. The band lies inside n77, so the radios exist and so does the release. What a carrier lacks is the licence, which Auction 115 sells from April 2027, and the clearing, which does not allow service in the top 75 areas until 31 December 2030. The handset base is missing too, and the order requires it by the same date. XR is the opposite case. Its specifications are frozen and its band is licensed, and what nobody has yet published is a device and a service that an operator sells.

That leaves the three rows that wait for Release 21, and each is waiting on something different. Sensing has a channel-model study from Release 19 and, in Release 20, a work item that covers base-station drone detection and nothing else. The 7 GHz radio has a federal report, a conference decision and an auction ahead of it, in that order, and the statute gives the auction until July 2033. The sub-millisecond and centimetre targets depend on both of the others, and there is a further wrinkle. They are finer than the 1 ms minimum the draft IMT-2030 requirements ask for, so a radio that meets the minimum need not deliver them. Put the three together and the earliest any of them has a specification to build to is March 2029, with a radio to follow. Every 6G revenue case inside the decade falls after that.

The rule never brings a date forward, and the last cycle moved four dates in five

The rule we have applied is a simple one: a capability earns on the day its last dependency arrives, and not a day before. Run it over the table and the earliest anything earns as 6G is 2030. New mid-band capacity earns from 31 December 2030. Everything a carrier earns before then is 5G-Advanced, on bands it already holds.

What makes the rule uncomfortable is that it only ever works in one direction. Nothing earns before its last dependency, so no date can come forward; a date can only go back, and it goes back every time a dependency slips. How often is that? In the 5G cycle, four releases in five (Exhibit 4). Release 15’s late drop and Release 18 each lost three months at a December plenary. Release 16 moved twice, six months in all. Release 17 moved twice as well, in March and December 2020, and froze on 18 March 2022, nine months after the date it was first scoped to. Only Release 19 held the date it had published, September 2025, set in April 2024.

Apply the rule to a roadmap and two lines appear. The 5G-Advanced line starts now. The 6G line starts around 2030, and if the last cycle is any guide it starts later rather than earlier. The next three years of revenue fall in the gap between the two, and the first six rows of Exhibit 3 are what fills it.

A roadmap should carry two lines and a capacity purchase between them

Set the questions on our 6G roadmap decision page against the table and each has a specific answer. The plan itself comes down to two lines with one purchase between them (Exhibit 5).

The decision page asks a handful of questions, and each one now has a place to look. Where the milestones fall is part 1’s table. Which capabilities carry a business case before the freeze is the first six rows of Exhibit 3, and what the spectrum pipeline releases, and when, is part 2. The dependencies that gate each bet are the three columns of Exhibit 2. The indicators that would move the timeline are the subject of part 7.

One thing a roadmap should not say is that any row below the sixth earns before 2030. Nothing on the record supports it, and a plan that depends on it is depending on the kind of date that held once in five in the last cycle.

The terms, briefly

• Release and freeze. A 3GPP release is a numbered set of specifications; its Stage 3 freeze is the point after which the radio protocols stop changing, and is what vendors build to. The later code freeze fixes the machine-readable encodings.
• Band n77. 3GPP’s name for 3.3 to 4.2 GHz as a time-division band. The C-band of 2021 and the upper C-band of 2027 are both inside it; a footnote restricts the American part of it to 3980 MHz until changed.
• RedCap. Reduced capability, a Release 17 device class with fewer antennas and narrower bandwidth, for wearables and sensors. eRedCap in Release 18 caps the peak rate at 10 Mbit/s to cut cost further.
• Partial economic area. One of 416 licence areas the FCC uses for mid-band auctions. The top 75 by population held 72.1% of Americans at the 2020 census and are the first to be cleared and served.
• Customer equipment. The fixed wireless router or outdoor unit a subscriber installs; for the fixed wireless row it is the device on sale.