AI startups are choosing .ai half as often

In the third quarter of 2024, 26% of AI products launching on Product Hunt registered a .ai domain. Two years later that figure is 12.2%. Over the same stretch those companies moved toward .com, from 42.0% to 47.8%.

The decline did not follow a bust. AI products held between 22.5% and 30.4% of all launches throughout, and total launch volume grew roughly sevenfold. The companies kept coming. A shrinking fraction of them bought the extension named after them.

The numbers

We recovered the web address of every Product Hunt launch between 1 September 2024 and 1 September 2026. Of 16,861 products, 14,354 had an address we could resolve, and 11,422 of those had registered a domain of their own rather than pointing at an app store or a hosting platform. Within that set, 3,170 are AI products by their own name, tagline or topics.

The early quarters are small, and 2024 Q3 rests on 50 launches. The trend does not depend on them. Take the last four quarters alone, each built on hundreds of products, and .ai still falls from 15.6% to 12.2% while .com climbs from 37.2% to 47.8%.

The same shape appears across all launches rather than only AI ones, measured month by month rather than quarterly. .ai was 8.8% of new domains in September 2024 and 3.6% in August 2026, the last full month in the set, declining in most months between and recovering in none of them.

The registry is not losing customers

A falling share is not a falling business, and the difference is worth being exact about, because the two get conflated every time somebody writes that an extension is dying.

In the third quarter of 2024, 13 AI launches here registered a .ai. In the second quarter of 2026, the last full quarter in the set, 142 did. That is eleven times as many, over a period when the share fell from 26.0% to 16.2%. The registry sold more every quarter. It simply sold to a smaller fraction of the companies in front of it, because the number of those companies grew faster than its own take.

So nothing here says .ai is in trouble. What it says is that .ai stopped being the obvious default for a category that used to treat it as one, and that is a statement about how founders now read the extension rather than about the registry's revenue.

What this is not

It is not a story about AI cooling off. The AI share of launches was 22.5% in the first quarter measured and 24.8% in the last, with a peak of 30.4% in between, so whatever happened to .ai happened while the category stayed large.

It is also not a measurement of the whole market. Product Hunt is where independent and early-stage products launch. A Series B company renaming itself does not show up here, and neither does anything with a procurement department. What this set captures well is the naming decision of small teams shipping something new, taken at the moment they ship it.

And we can show the movement without explaining it. Three explanations fit the data equally well. A .ai renewal costs several times a .com and arrives annually, which is a different commitment from a one-off purchase. The short, obvious .ai names went early in the boom, so a founder arriving in 2026 finds a worse selection than one arriving in 2024. An extension that announces a category is an asset while the category is novel and a liability once it is ordinary, and AI stopped being novel somewhere in this window. Our data cannot separate any of these three from the others, and we would not want a reader to take our ordering of them as a ranking.

Why it matters beyond .ai

.ai had every advantage an alternative extension can have. Its category was the defining technology story of the period. The meaning was exact, requiring no explanation to anyone. Awareness was total among the people buying domains.

It still lost more than half its share inside its own category, during that category's peak, to .com.

That is the general finding. An alternative TLD does not absorb demand from .com so much as borrow it while the association is fresh, and the loan comes due. .xyz did this. .io did a gentler version of it, and sits at 4.4% of launches here. The pattern is old enough that .ai's turn was predictable in shape even if the timing was not.

For anyone naming something now, the practical reading is narrower than the thesis. The extension you pick because it means something today is the extension that will date your product. .com means nothing, which is exactly why it does not expire.

The launches come from the public Product Hunt API. The full method, the decisions that could have gone another way, and the dataset as a CSV are on how we measured this.

The shape of the rest

For completeness, here is what the 11,422 registered domains actually used. The long tail is longer than it looks: 342 distinct extensions appear.

.com takes not quite half. The second-placed extension takes less than a sixth of what .com does. Below .dev the field fragments into hundreds of extensions with a few dozen launches each, which is the same fragmentation that makes any one of them a weak signal to a visitor.