Naming for New gTLDs: Names That Read Across the Dot

Brandable domain names on new endings last when the ending does work inside the name, so the whole thing reads as one phrase across the dot. But when the ending is only a cheaper stand-in for .com, owners tend to stop renewing. The Domain Name Industry Brief for Q3 2026 puts the combined new gTLD renewal estimate at 28.6%, against 79.8% for other legacy gTLDs (DNIB). Below are the numbers, our estimate of when the internet reaches a billion domains, the strings the 2026 round adds, four naming patterns, and a five-test scorecard you can run on any candidate.

What the Q3 2026 Domain Name Industry Brief says

The Q3 2026 report from DNIB.com (sponsored by Verisign) is the best single snapshot of who is buying what. And the figures at 30 September 2026 are the useful starting point:

• All TLDs: 411.6 million registrations, up 9.9 million (2.5%) on Q2 and up 33.0 million (8.7%) year over year.
• .com and .net: .com at 169.8 million.net at 12.6 million. The preliminary combined renewal figure for the pair is 74.3%.
• ccTLDs: 151.4 million, up 4.6% year over year. DNIB says the available data is not enough to estimate a combined ccTLD renewal figure.
• New gTLDs: 56.6 million, up 13.7 million (32.0%) year over year, which is 13.7% of all registrations. The combined renewal estimate is 28.6%.
• Other legacy gTLDs (everything except .com and .net): a combined renewal estimate of 79.8%.
• Root zone: 1,122 gTLDs delegated.

The top ten gTLDs are .com.net.org.xyz.top.info.shop.online.store and .vip. Inside that one list, renewal estimates run from 86.3% for .org to 12.5% for .top. Same table, same quarter, and a gap of roughly 74 percentage points. That is not noise, it is the market telling you how differently these names behave.

The point is uncomfortable. New endings get bought fast and kept rarely. The new gTLD base grew by nearly a third in a year, and fewer than three in ten of those names renewed at the latest estimate. Verisign's own summary of the report carries the same headline totals, if you want a second place to check them.

When the internet reaches 500 million and one billion domains

Some context for how fast the base is moving. Total registrations at 30 September were 362.3 million in 2024 (up 0.8% year over year), 378.5 million in 2025 (up 4.5%) and 411.6 million in 2026 (up 8.7%). Annual additions went from 3.0 million to 16.2 million to 33.0 million. So yes, that is a steep climb in two years.

Our estimate. We think the internet passes 500 million domains during 2029, most likely in the first half, and that it does not reach one billion before the late 2030s, with the 2040s the more likely decade. This is a projection from three data points. It is not a forecast from DNIB, Verisign or ICANN, and it should be read as arithmetic with a point of view attached.

The scenarios, all from the Q3 2026 base of 411.6 million:

Why do we lean towards the slower end for one billion? Two reasons, in plain words. First, the recent growth is led by new gTLDs, which supplied 13.7 million of the 33.0 million year-over-year increase and renew at 28.6%. A lot of each year's growth leaves again a year later, so a fast year does not carry forward the way it would if everything renewed. Second, the 2026 round adds new endings from late 2027 or early 2028. That adds supply, and supply is not the same thing as names people keep.

The 500 million call is firmer because it only needs the next couple of years to look roughly like the last one. The billion call needs a decade of assumptions. So treat it with suspicion, ours included.

Why most new-ending names do not last

Think about what a renewal invoice actually asks. Is this still the name? If someone registered brand on a new ending because the .com was taken, the ending is a placeholder. A year later the invoice arrives, nothing about the ending has earned its place, and the placeholder gets dropped. But if the name was built so that the ending is one half of a phrase, dropping it means dropping the brand. That is a harder decision, and far fewer people make it.

I can't read minds from a DNIB table. The report gives renewal percentages, not reasons. But the shape of the data fits the idea. The new gTLD renewal estimate was 32.2% in Q3 2025 and 31.0% in Q2 2026, and it is 28.6% now. Over the same stretch the base grew from 42.9 million to 56.6 million. Bigger and less sticky at once. Growth by volume and retention are two separate things.

And the ending alone does not settle it. .org and .top sit in the same top ten with renewal estimates of 86.3% and 12.5%. That is a fact about how names on those endings are used and renewed, not a verdict on either one.

There is a counter-position worth reading. A 2014 CircleID post argued that new gTLDs in general (except, perhaps, any offering outright free names) presented a strong case for renewal. The 2026 combined estimate is a different picture. One caution cuts the other way too: Verisign attributed a small dip in .com and .net renewal (74.3% expected against 75.4% a year earlier) to the mix of first-time and previously renewed names, according to its Q3 2026 earnings call. A young base renews differently from an old one, and new endings are young.

A blunt test sits underneath all of this. If this exact name appeared on .com tomorrow, would you move? If yes, the ending is a stand-in. If you would shrug and keep the original, the ending is doing its job.

What the 2026 round adds as naming material

ICANN published the applications on Reveal Day, 7 October 2026. The announcement gives 1,615 applications, of which 333 are brand applications. Those brand endings will never be sold to the public (more in our post on .brand TLDs). From the Reveal Day contention data: 481 applicants, and 891 applications (55%) sitting in 263 contention sets, against 39% in 2012. Our full breakdown of the round has the wider picture. The snapshot is as of 7 October, and ICANN says the final numbers come on String Confirmation Day.

The most contested strings, by number of applicants:

• .agent: 13
• .bit: 11
• .api.brand and .hub: 10 each
• .pop and .robot: 9 each

Further down, other word strings with several applicants: .vibe has 7.human 6, and .lab.mind.ask.key and .therapy 7 each.

Then there are the single-applicant words. Squarespace Registry applied for 15 small-business words, including .gym.chef.venue.learn and .skill. And by webhosting.today's count from ICANN's contention file [their number, not ICANN's], about 408 generic strings had a single applicant, for example .vps.cdn and .saas.

One moving part matters. Replacement-string swaps closed on 21 October and changed some sets: Link Freedom Group moved from .brand to .industry. ICANN publishes the final list on String Confirmation Day, 17 November 2026, so treat every count here as a Reveal Day figure.

Four naming patterns that read across the dot

Every example below is built only from strings in the round, and every one is illustrative. None is a claim that the name is available, registered or free of trademark problems.

Phrase across the dot

The label and the ending together say something a person would say anyway. slow.chef , iron.gym , quiet.venue. Read it with a space instead of a dot and it still makes sense. This is the strongest pattern because the ending carries meaning that the label does not repeat.

Action plus ending

The label is a verb or an invitation and the ending completes the object. start.ask , join.hub , go.learn. These read well aloud and suit products with a clear call to action. But the risk is blandness: if the verb is generic, lots of people will think of it.

Category signal

The ending says what the business is and the label says which one. travel.agent , weather.api. The signal is immediate, which is also the catch. Descriptive combinations are usually harder to protect and easier to confuse with each other, and .agent drew 13 applicants for exactly this kind of reason.

Hacks where the ending completes a word

Domain hacks split a word across the dot. ha.bit reads as habit, re.mind as remind, mon.key as monkey. Some people enjoy these puzzles. Searches for domain hacks are small but steady, and the pattern is old. The problem is speech. Say ha dot bit to someone on the phone and they will type something else. A hack can be clever and still fail the moment it leaves the screen.

For endings you can register today, our post on fun domain extensions covers the same idea with live options.

The across-the-dot scorecard and the strings table

Five pass/fail tests. Run them on a candidate before you fall in love with it.

1. Reads as one phrase across the dot. Swap the dot for a space. If a stranger would parse it as a single idea, it passes.
2. Survives being said aloud. Say it to someone, no spelling allowed. If they type it correctly first time, it passes. Most hacks fail here.
3. Still works as a name without the ending. Drop the ending and keep the label plus its meaning. If you could still brand with it, it passes.
4. Can be trademarked. Search the relevant registers for the full name and for the label alone. Descriptive combinations are usually harder to register. Our post on trademark or domain first covers the order of operations.
5. The ending is not contested or brand-only. If the string has several applicants, your launch date is uncertain. If it is one of the 333 brand applications, it will never be sold to you.

Five passes: go. Pass 1 to 3 but fail 5: wait, because the name is fine and the ending is not ready. Fail any of the first three: rename. A business name generator with a trademark check is a quick way to catch test 4 failures before you get attached to a name.

Contention is not a verdict on a string. It tells you about timing and uncertainty, nothing more.

And the strings, grouped by naming pattern. Applicant counts are as of Reveal Day, 7 October 2026. Examples above are illustrative.

The availability column leans on ICANN's own wording that it had not yet set start dates for contention resolution and later milestones. Nothing here predicts which application will succeed.

What to do now

None of these endings can be registered yet. The timeline from ICANN's announcement:

• 17 November 2026: String Confirmation Day, the start of the Community Input and Objections Period.
• 17 November 2026 to 16 March 2027: community input and objections. WIPO handles Legal Rights Objections and String Confusion Objections, which is one way sets can shift.
• First half of 2027: the Prioritisation Draw, which sets processing order. The String Evaluation Stage is expected to last 180 days, and processing of applications and contention sets does not start until it ends.
• Late 2027 or early 2028: the first uncontested TLDs go live. Contested strings come later.

So the practical advice is boring on purpose. Pick a name that works on an ending you can register today. Then claim the 2026 ending when it opens: at sunrise if you hold the trademark, or at general availability if you do not. A name that only works on an ending that does not exist yet is a bet on a launch date nobody has set. If you are still at the start, our brand naming process covers how to build the name before you pick the ending.

One practical risk remains. Some software and email forms still reject long or new endings, a problem known as Universal Acceptance. Before you print a new ending on a van, a menu or a business card, try it in the contact forms, checkout flows and email fields your customers will actually meet.

FAQ

Can I register a .agent domain yet?

No. A .agent domain cannot be registered today. The string has 13 applicants at Reveal Day, the most of any string in the round. Expect it after the first uncontested launches, and expect no date until ICANN sets contention resolution timing.

When will there be a billion domains?

Not soon. Our estimate puts 500 million during 2029 and one billion in the 2040s more likely than the 2030s. The scenario table above shows the range, from around 2037 to around 2060. Those are our projections from three data points, not an ICANN or DNIB forecast.

Are domain hacks a good idea for a brand?

They can look clever, but they fail the spoken test more often than any other pattern. If customers have to be told how it is spelt, you pay for that on every phone call and podcast mention. Use a hack only if the split falls somewhere people would guess.

Can I buy a name on one of the 333 brand endings?

No. Brand applications are for the applicant's own use and will not be sold to the public. Our post on .brand TLDs explains how they work.

The sharpest caveat in all of this is the renewal gap: 28.6% for new gTLDs against 79.8% for other legacy gTLDs, so an ending needs a reason to exist inside the name. If a candidate fails any of the first three scorecard tests, it is not ready for any ending. When you want a second pair of eyes on a shortlist, a naming consultant can run the five tests with you.