LLL.ai: The Run to Zero, and What .ai Sales Did Meanwhile

Part 1: The supply curve

There are 17,576 possible three-letter combinations. By September 2025, 91 percent of them were already registered in .ai. We put the remaining 1,424 on a list and rechecked every name against registrar availability data every six hours, dropping each one the moment it came back as registered. The log kept every count and every removal, so the whole draw-down can be replayed.

How it went

Roughly half the supply went in the first four months. The first week alone took over 200 names, and one burst on 21 September accounted for 40 of them. That burst was us: we registered 41 names in that round, six more in October, and fifteen in July 2026, on top of the 34 we already held from before the tracking started.

From March the pace held near four names a day. Then in August the tail was swept. The last 47 names went inside eighteen hours on 30 August, and a whois check shows most of them created within the same second at a single registrar. Someone bought the remainder in one go, and the odds are it was an investor rather than 47 startups arriving at once.

What was left at the end were the hardest combinations. Across the final 200 names taken, j, q and z were by far the most common letters. The names with vowels and pronounceable shapes had gone a year earlier.

The live page with the full replay is at domaincracy.com/services/availability/lll-ai.

Part 2: What .ai sold for in the same twelve months

Supply running out is only half a story. The other half is what buyers paid. We compared the reported .ai retail sales we track for September 2025 to August 2026 against the twelve months before. These are public sale reports, so they cover the visible market only, and reports for the most recent months are still arriving. Treat the last quarter as incomplete.

Read together: the .ai aftermarket did not accelerate while the LLL supply ran out, but it did not cool either. Volume and count were flat to slightly up, the median slipped a little, and the top end produced the first reported seven-figure .ai sale, bot.ai at $1.2M in February 2026. Below it, single dictionary words did the heavy lifting: wisdom.ai at $750,000, lotus.ai at $400,000, and a band of names at $100,000 to $165,000 including speed, flourish, vesta, deep, amber, certify, odds, confidential, climb, dealer and enclave.

Against .com, the king

Flat can mean stalled or it can mean stable. The way to tell is to put .ai next to .com over the same two windows, from the same set of reported sales. The .com count per month held near 250 through May 2026 and the .com median stayed pinned at $5,000 in both years, so the reporting base is comparable. What changed in .com was the upper tier: sales at $25,000 or more fell from 245 to 167, and sales at $100,000 or more from 57 to 36. Brokered .com sales tend to be reported late, so part of that gap will close, but not all of it.

Our reading: .ai is stabilising rather than stalling. While the .com upper tier thinned in the reported data, .ai held its count, its volume and its share of six-figure sales, and it is no longer a segment carried by one or two headline deals a year. A median near $15,000 across nearly 300 reported sales is a market with a middle, which .ai did not have three years ago.

The three-letter .com row is the more interesting one for what comes next. That class has had zero supply for almost thirty years, and in the past twelve months it cleared at a median of $112,500 with half the reported sales above $100,000: hsm.com at $550,000, txt.com at $502,250, pub.com at $500,000, kbf.com at $260,000, and a run of random-letter names such as ocf, mpa, ino, yal, zmo and nuq between $95,000 and $170,000. Nobody claims three-letter .ai will reach .com levels. The point is what a sold-out class looks like once buyers have no registration alternative: the random-letter tier stops being priced against a registration fee and starts being priced against the next seller.

Three-letter .ai specifically

The LLL.ai slice is small, so the numbers are indicative rather than statistical. We track 38 reported LLL.ai sales in the earlier window and 27 in the later one, with medians of $19,065 and $17,000. Two tiers are visible in both years:

• Word and acronym names sold at $45,000 to $100,000 during the run-out: evo.ai and hio.ai and adg.ai in one January batch, lio.ai, una.ai, ado.ai, pcc.ai, cui.ai. Above them sits bot.ai on its own.
• Random-letter names, the same kind that were still available to hand-register until August, cleared at $3,000 to $8,000: gxl, jov, jez, gxi, fso, ffo, cth, crq, kmk. That is many times a registration, but it did not move upward during the year.

So the honest reading of our "new floor" line from a year ago is this: the floor did not rise while supply was still trickling. Every buyer who wanted a random LLL.ai could still register one, and the aftermarket priced accordingly. That option closed on 30 August 2026. From here, the only way to get a three-letter .ai is to buy one from whoever holds it.

The first post-zero asks are already visible. Within two weeks of the sweep, names from the final batch such as nzj.ai and ozj.ai were listed at $4,700 each on the registrar's own sale pages. That sits inside the $3,000 to $8,000 band the random-letter tier cleared at over the past year, so the buyer of the tail is pricing to the existing market rather than above it, at least to start. Forty-seven names, $4,700 each. Make of the coincidence what you will. Whether the tier lifts from there is the question for the next twelve months, and we will run this comparison again when the reports for the coming quarters are in.

Our position

We hold 96 three-letter .ai names, picked in three rounds as the pool thinned. They are listed on our portfolio page. We are not in a hurry: the point of the chart above is that the pool is gone, and a finite pool with steady demand tends to reprice slowly and then all at once.

Method

Supply: the list started from every three-letter .ai name that showed as unregistered on 13 September 2025, rechecked every six hours against registrar availability data; a name left the list the first time it came back as registered. Our checker was offline from 21 January to 9 March 2026; names taken in that window are counted on the day it resumed and drawn as an interpolated stretch in the chart. Sales: reported retail sales we track from public sources, nominal prices, dated sales only; 132 further 2026 sales in our set carry a year but no date and are excluded from the monthly figures. Registry totals and undisclosed private sales are not included.