The Super-Intelligence Rush: who gets to a name first, and why

On 19 September, at 15:29 UTC, President Trump posted a poll on Truth Social asking which name should replace "artificial intelligence": Superior Intelligence, Extreme Intelligence or Supreme Intelligence. Two of the three options abbreviated to SI, and new .si registrations rose from 143 that day to 928 the next. A second poll on 20 September narrowed the field, a third on 21 September offered "super intelligence" for the first time, and on 22 September he told the United Nations General Assembly that the United States would call AI "super intelligence," or SI. On 29 September an executive order made the term official across the federal executive branch. Within two weeks of the first poll, a 34-year-old country-code domain that belongs to Slovenia went from about a hundred new registrations a day to more than eleven thousand.

The registry's own published series shows the shape of it. New registrations reached 4,478 on 23 September, the day after the speech, and from 30 September through 4 October the daily count stayed above 11,000, peaking at 13,651 on 4 October. In ten of the twelve months before September 2026, new .si registrations ran between roughly 2,300 and 4,000 a month, and even the busiest of them, June 2026, brought 6,938. The whole of 2025 brought 29,410, and September 2026 alone brought 46,066 new registrations, more than one and a half times the previous year's total. The registry's own fee did not change: it has charged registrars €10 a year, excluding VAT, since 1 January 2023, and what an owner pays is set by each registrar. What changed was the story attached to two letters.

What the rush actually bought

We wanted to know whether the rush was broad or targeted, so we drew a neutral sample of 180 strings: 120 four-letter strings built on consonant-vowel patterns by a generator with a fixed seed, and a fixed list of 60 common English nouns of four to six letters. The sample was set in code before this run's first query, and we publish the generator and seed with the data; 35 of the 60 words had already been looked up in an earlier check that day. We then read all 180 directly from the .si registry on 5 October. Across 2026, monthly registrations in the sample ran between zero and five until September, then rose to 15 in September, 13 of them after the first poll, and 16 in the first four days of October. One of the October names was already in its redemption period when we read it, so that count may fall by one.

Of the sample strings still unregistered at the end of August, we estimate that about 45 percent were registered by 4 October: 31 of 67, or 29 of 65 if only registrations after the first poll are counted. The estimate assumes that no name in the sample was deleted and registered again during the window, and 18 of the 31 went through a single large registrar. The common words tell their own story, because 57 of the 60 were already registered before September and the three still open were all taken during the month. In our sample, then, the rush looks broad rather than targeted, and most of what it took were short four-letter strings, the part of the space that was still open.

Independent trackers saw the same shift toward short, letters-only names from the outside. Netcraft reported that just over half, about 52 percent, of the .si names it detected as registered at one large registrar between the first poll and the speech were listed on that registrar's own aftermarket by 23 September. DomainKits found that names of five characters or fewer rose from 18.8 percent of the new .si names it observed from 1 September up to the first post to 34.1 percent during the wave. It also found that 16.2 percent of the names registered during the wave pointed their nameservers at large sale platforms, up from 4.3 percent before it; that is a floor rather than a full count, since names offered for sale through a registrar's default settings are not included.

The wave also copied strings across extensions. Netcraft counted more than 44,000 .si names, about 22 percent of the whole zone, whose string also exists in .ai, but most of that overlap is older than the rush: the registry logged only 8,231 new .si names from 19 to 23 September. The copying inside the wave shows up in DomainKits' sample instead. Of the names registered during the wave, 98.4 percent were also registered as .com, against 66.2 percent before it; 48.5 percent existed in more than 50 extensions, against 23.3 percent; and only 1.2 percent existed in .si alone, against 28.8 percent.

Supply widened as demand arrived. Two new sales channels began offering .si on 23 September, the day after the speech: a large web host reselling the extension, and a newly accredited registrar whose start Netcraft dates to that afternoon. The first surge, on 22 and 23 September, ran largely through a registrar that was already selling .si, and most of the new channels' volume came after the 29 September order.

The question underneath: who saw it first

Every rush raises the same suspicion among the people who arrive second. An owner searches for a name, hesitates, and finds it gone days later, now parked on a marketplace at many times the registration fee. The suspicion has a name, front-running, and ICANN's own Security and Stability Advisory Committee named it in October 2007. The committee warned that checking whether a name is available can itself reveal an interest in it, or the value someone places on it.

In January 2008 a major registrar began reserving every available name searched on its own site for four days, which meant that during the hold the name could be bought only there. It described the hold as protection against front-runners; a class action filed that February called it a monopoly on searches. What followed was tested, and the tests did not find front-running. When the advisory committee reviewed about 120 complaints in early 2008, it could prove none of them and concluded that the perception was broader than the reality. A study commissioned by ICANN, announced in 2009, ran more than 600 test searches over ten months and found no evidence. A 2010 university study searched 16,635 available names at the 19 largest registrars, held back 43,629 as a control, and found no registrar linked to a significant rise in registrations of the names searched; its authors warned that a registrar could still hide the practice by registering through affiliated companies. Registrars have denied front-running throughout, and the public accusations made since then have stayed anecdotal.

We checked our own searches against the record before saying anything. On 5 October, between 08:56 and 09:23 UTC, nineteen .si names were searched in the ordinary course of our work, all through our registry-direct tool, and afterwards we read each one again at the registry. Fourteen came back registered, and every one of them had been registered before it was searched: nine during the rush, through five different registrars, the most recent at 05:18 UTC that morning, four hours before our search; and five earlier, one in 2008, one in October 2025 and three in late May 2026. The other five were unregistered when we searched them. As of 9 October, four of those five are still unregistered. One, lande.si, was registered on 9 October, four days after it was searched. We are keeping the watch on the other four through 19 October and will publish the result.

What this supports is narrower than a verdict. Our 5 October searches came after every one of the fourteen registrations, so in this case the rush got to those names first and needed no one's search to do it. The one registration that followed a search, lande.si, is a single name four days into a rush that was adding more than eleven thousand names a day, and it cannot be told apart from the sweep. Whether earlier searches made elsewhere, in a registrar's search box or on a lookup site, played any part is a question our own records cannot answer, and it stays open. That is a simpler finding than a conspiracy, and in some ways a harder one, because a market that sweeps the open space does not need to watch anyone.

That structure is the real subject. The search box at a registrar is a live feed of what owners want, captured by a company that also sells names. No ICANN rule forbids a registrar from acting on what people search for: ICANN bans front-running by registries, not registrars, and the .si registry's contract with its registrars is silent on it. At least one country-code registry, .au, has restricted its registrars from registering names for themselves since at least 2008, so the norm can be written. Registrar privacy policies, meanwhile, are written broadly. Two of the six major registrars whose policies we read list search history among the data they collect and may draw on for marketing or share with partners, and none of the six promises to keep domain searches out of the registrar's own decisions. An owner cannot see what happens to that feed or audit it, and the only check available from outside is a test.

A test anyone can run, and an alternative that exists

The method that settles the question is not complicated, and it is essentially the design of the 2010 study. Choose twenty unregistered names that a real buyer would want. Search ten of them only in a registrar's own search box and the other ten only through a tool that reads the registry directly, register none of them, and watch all twenty at the registry for fourteen days with every registration dated. If the first group disappears and the second does not, the record speaks for itself; if neither moves, the suspicion was misplaced for that registrar in that window. Either way, the result is evidence instead of anger, and our own five-name watch is only the registry-direct half of that design.

The alternative to the feed already exists. A search can go straight to the registry that keeps the authoritative record, with nothing stored, nothing passed to a registrar, and nothing sold. That is how MetaTaping reads a .si name, and it is the practice behind the standard we hold all of our search surfaces to, which we call Observer Separation: a search is never a purchase signal, and whoever watches what people look for must never become the undisclosed trader against it. We set the structure out in public because we think the wider market needs a norm that only parts of it have.

Our working hypothesis is that the super-intelligence rush will cool and that most of the swept names will never sell. The registry's daily series and the resale listings will test that, and we will update this piece as they move. What the rush has already left behind is a clearer picture of who gets to a name first and why: the party that sees the signal earliest. In a market where the signal is your own search, the safest search is one that nobody but the registry ever sees.