The land rush closed. Here is the board.

251 million names already contain the answer

You cannot poll the world about TLDs it cannot buy yet. But you can watch what people settle for. Every acmelabs.com is an acme.labs that never had the option; every paymenthub.io is a payment.hub. So we took the full zone index, 382.9 million registered domains deduplicated to 251.75 million unique names, and counted, for 683 candidate and comparison strings, how many registrations end in each one. Registrations, not unique strings: a name registered in forty TLDs is forty proofs of demand, and every count and model figure on this page is on the registrations basis. This is the same method our .web forecast used in July; this scan reproduced that report’s audited numbers within half a percent.

Then we did the part that separates signal from artifact. Raw ending-counts lie. Two hundred thousand names end in “king”, but booking, parking, and ranking are not aspiring royalty; they are English words that happen to end that way. For every string we sampled recent registrations and measured what fraction are genuine compounds, a real prefix attached to the word, versus dictionary coincidence. We call it purity, and it moves everything: it cuts “king” by nine-tenths and leaves “hub” nearly whole. Fail to do this and you will pay $227,000 for a mirage.

The string sets the ceiling. The launch picks the floor.

Here is the part that changed how we rank everything. We took 39 strings that already became TLDs, measured their purity-adjusted ending demand the same way, and divided each TLD’s actual zone size by it. That ratio is the conversion rate: how much of a string’s measurable demand the registry actually captured. It does not cluster by how good the string is. It clusters by how the registry behaved.

Read that chart the way an operator would. A .group-quality string with .group-quality distribution converts twenty percent of its demand, reliably. The same string with a funded launch converts five times more. Google-grade execution, twenty times. Nothing about the string itself changed. Whoever tells you a string is intrinsically worth eight figures is selling you the ceiling; whoever budgets only the application fee is buying the floor.

Now hold every string to a 5-year payback

The economics of running a TLD are almost embarrassingly small: about $100k a year keeps one alive (ICANN fees, a backend provider, escrow, abuse handling), plus whatever you spend on marketing, which section 3 just told you is the whole game. At $15 wholesale a TLD breaks even for good at roughly 12,400 names. At $50, about 4,100. Renewal data settles the pricing debate before it starts: standard-priced registries renew at 55 to 83 percent, teaser-priced ones at 3 to 30. Price like .app, never like .icu.

We modeled all 200 finalists at their recommended wholesale price, with a four-year zone ramp, premium programs, and each string’s expected auction cost baked into the capital. Then we asked the question a five-year investor asks: when is the money back?

Every string that passes the five-year test, priced

All 182 strings whose funded-launch case returns the capital within five years, ranked. Zone and 5-year net are the funded case at the recommended wholesale price; payback is shown for both cases. Auction column is our estimate of what winning a contested set costs, from 2012 comps and announced applicants. Risk dots: green = low or none, amber = medium, red = high or extreme. A dash in the shelf column means it never pays back within ten years without marketing.

The first 341 strings are public

Two applicants published their lists before ICANN does. Link Freedom Group, whose Nova Registry runs .link and its 288,000 names, filed 316 strings. Oinkadot, formed by Ray King of Porkbun and Todd Han of Dynadot, filed 25. Both lists are measured here exactly like the board above: net suffix demand, purity adjusted, converted to a mature zone. The only change is the cost stack. A registry running dozens of TLDs off one platform breaks even near 4,700 names at $15 wholesale, where a standalone registry with its own marketing budget needs 12,400.

What registry premium names are actually worth

Every model above prices a TLD on ordinary registrations. Registries also reserve their best names into premium tiers that sell for hundreds or thousands of dollars a year, and renew at that price. For a small zone this is the difference between a slow loss and a small profit, so it deserves real calibration rather than a rule of thumb. Radix is the only portfolio operator that publishes the detail, so it anchors the numbers.

Every assumption on this page, as a dial

The whole report rests on a handful of numbers: how much suffix demand converts into a zone, what you charge, whether you run a premium tier, and what an auction costs. Pick any of the 486 strings we modelled, including every string the two applicants filed for, and move them. The starting values are the ones used in the tables above.

Portfolios, because single applications are coin flips

Contention risk is binary per string and diversifiable across strings. Reveal Day resolves it all at once. These three baskets were built from the board above before the window closed, with 5-year numbers under the funded-launch case and every auction estimate included in the capital. They are worth keeping on the page now for a different reason: the filings ran straight through them. Link Freedom took seven of the Cashflow Eight, eight of the Conviction Ten and six of the AI-Era Ten. What survives untouched is listed under each basket.

The auction history everyone should memorize first

Seventeen strings reached ICANN’s last-resort auction in the 2012 round; a couple hundred more settled privately. The distribution is the budgeting tool: a median contested generic cost $1M to $3M, strong commercial keywords $3M to $7M, category-definers $19M to $41.5M, and one strategic outlier went for $135M. This round the private exit is gone. Whatever you bid, you either win or burn the legal spend, so the walk-away number belongs in writing before Reveal, not after.

Falsifiers, then footnotes