Why All the Good AI Domain Names Are Taken (and What to Do About It)

The Land Grab, By the Numbers

Start with the plainest evidence: how fast .ai actually grew. This isn't modeled or estimated — it's the registry's own reported counts, corroborated across multiple domain-industry trackers and Anguilla's own government communications.

Sources: Domain Name Wire (Oct 2024 deal), Anguilla Focus (1M milestone).

Anguilla's Windfall

Every one of those registrations is a transaction with a country of about 16,000 people. Government revenue from .ai domain sales rose from roughly $32 million in 2023, to about $39 million in 2024, to approximately $85.3 million in 2025 (figures converted to US dollars) — and a government official told BBC Radio 5 Live the 2025 figure was nearing half of Anguilla's total government revenue that year. The 2024 outcome blew past its own budget: "Domain name registration is projected to exceed the 2024 Budget estimate of $64.05m by an impressive $41.48m, therefore exceeding $100m," Anguilla's Premier Ellis Webster said at the time, in Eastern Caribbean dollars (Anguilla Focus) — a two-letter suffix now sits among the most consequential line items in a national budget.

Sources: Anguilla Focus (2024), Anguilla Focus (2025), IMF.

The 2025 Auction Overhaul

The .ai registry itself changed hands mid-boom. In October 2024, Anguilla's government signed a five-year deal handing day-to-day technical management to Identity Digital — the same company that runs .io's backend. The actual technical transition completed on February 25, 2025. Identity Digital immediately overhauled how expired and abandoned .ai names get resold: monthly auctions through Dynadot became daily auctions through Namecheap, and names that fail to sell now feed into Identity Digital's DropZone system instead of dropping and disappearing outright. The effect showed up fast — over $600,000 in expired-domain auction sales in the first month alone, plus a reported 60% year-over-year jump in regular, non-auction registration revenue (Domain Incite). Practically, this means the supply of "used" .ai names now churns faster than it used to — good news if you're patient, since names do come back. They just come back through a bid, not a free grab.

What AI Domain Names Actually Sell For

Headlines love the extreme end of this market. Here's the actual documented ceiling, cross-checked across multiple reports, followed by what's typical.

Sources: DNJournal (ai.com $70M), Axios (chat.com), TechCrunch (ai.com history), AOL (you.ai), Domain Name Wire (aftermarket comps).

That table is the exception, not the rule — it's what gets written about, not what most people pay. NameBio tracked roughly 190,300 domain sales in 2025 totaling more than $244 million, up 31.9% in dollar volume over 2024. .com still took 72.0% of that volume; ccTLDs, including .ai, took 16.3% — up from 13.4% the year before, almost entirely on the strength of AI-related names (NamePros, via NameBio data). Below the headline sales, real keyword-AI .com transactions mostly land in the low-to-mid five figures: iqai.com sold for $44,442, hyperai.com for $40,000, cosmoai.com for $25,000, soulai.com for $21,000, with a long tail down into the low four figures. If your budget is realistic, plan around this tier — not the $70 million outlier at the top of the chart.

Why It Feels Like Zero

Two separate scarcity problems are stacking on top of each other. First, .com itself: Verisign's own registry data puts .com at a new all-time high of 163.6 million registrations, out of 392.5 million domains across every extension combined — and newer extensions are growing roughly 8x faster than .com in percentage terms, which tells you where the open names actually sit today (see our full breakdown of how many .com domains are registered). .com's absolute count keeps climbing because renewals still outnumber non-renewals, but the pool of clean, unregistered short words emptied out years ago.

Second, .ai: with roughly 2,000 new .ai registrations landing every day in early 2026, any name simple enough to guess quickly is also simple enough for someone else to have already registered — often within the same week you thought of it. That's illustrative math, not a verified per-name count, but it matches the felt experience precisely: it's not that AI names are impossible to find, it's that the pool of easy, obvious ones drains in days, not years. This dynamic is a real part of why .ai overtook .io among recent YC startups — and why the .ai vs .com decision increasingly comes down to category fit rather than which extension has more names left.

The Generated-List Trap

This is where most naming tools quietly fail: they generate names faster than they verify them. Type a business idea into a typical AI naming tool and you'll get a list in two seconds — a list that was never checked against a live registry, because checking is slower and less impressive than generating. Half the names on it are already taken. You won't know which half until you manually search each one.

It gets stranger. Researchers at Palo Alto Networks' Unit 42 ran 685,339 prompts across two major LLM families for 913 real brands and got back 2.1 million suggested URLs — of which roughly 250,000 were hallucinated domains nobody had registered yet, sitting open for anyone to grab. Attackers are doing exactly that: Unit 42 documented real phishing and malware campaigns active since July 2026 built entirely around registering a domain an AI model kept hallucinating, in some cases 18 to 51 days before attackers actually used it (The Hacker News). The lesson cuts both ways for naming a real business: a generated list can't tell you a name is available, and it can't tell you a name is safe. The only fix is checking live, every time — the entire premise behind our own AI domain name generator, and why every rescue path in our domain-taken guide starts with a real-time RDAP check, not a guess.

What Actually Works: Evidence-Based Strategies

None of the scarcity above means you're stuck. Four approaches consistently work, roughly in order of how much they're worth trying first.

1. Go Invented, Not Descriptive

The best-funded AI companies of 2025-2026 didn't win the naming game by finding a leftover keyword domain — they picked names that mean nothing on their own: Anthropic, Perplexity, Mistral, Sierra, Harvey. Y Combinator's own Fall 2025 batch shows the same instinct at the earliest stage, favoring short, abstract names over descriptive ones. Invented names have a practical advantage that has nothing to do with taste: the pool of possible short, pronounceable nonsense words is enormous and barely touched, while the pool of plain-English AI-keyword phrases is nearly exhausted. See our deeper breakdown in invented names vs. keyword domains for the data on why this also holds up better for trademarks and SEO.

2. Use a Modifier, Deliberately

When the bare keyword is gone, a prefix or suffix pattern — get, try, use, go, hq, join, meet — is the most reliable fallback, and it's why you see it everywhere in AI tooling. It's less a compromise than its own convention now: a modifier signals "early, direct, product-focused" the same way a bare .com signals "incumbent." The catch is legal, not aesthetic — bolting a modifier onto a live AI model or product name (GPT-anything, Claude-anything, Gemini-anything) can create real trademark exposure regardless of whether the domain itself is available. Check the underlying name first; our guide to AI model name trademark risk covers exactly which patterns are dangerous.

3. Pick the Alternative TLD for Fit, Not Desperation

"It was still available" is a worse reason to choose an extension than "my product is genuinely AI-native and my audience already reads .ai as a category signal." The same logic applies to .io, .dev, .app, or any other newer extension: they solve availability, but only the right one also does branding work for free. Pick backwards from the wrong reason and you'll spend years renewing a mismatched domain. Our .ai vs .com comparison walks through the actual tradeoffs — cost, trust, SEO — rather than just "is it available," which any live search already answers.

4. Buy on the Aftermarket, Properly Escrowed

If a specific name matters enough, buying an already-registered one is legitimate — and safer than most people assume, if you follow the size-based playbook the aftermarket already runs on: under $5,000, a platform's built-in buyer protection (Afternic, Sedo) is usually enough. From $5,000 to $25,000, use the platform's own escrow. Above $25,000, use a dedicated service like Escrow.com, where fees typically run 0.89%-3.25%, and funds are held until the transfer is verified on both ends. Never wire money directly to a seller, and never release funds before the domain transfer is confirmed registrar-side. It's slower than registering a fresh name — five to seven days is typical — but it's the one strategy that gets you the exact name you wanted, instead of the best name still open.

Frequently Asked Questions

Sources

Anguilla Focus — .ai crosses 1 million registrations (registration growth, daily registration rate)
Domain Name Wire — Identity Digital's October 2024 deal (500K registrations at the time, 20% of government revenue)
Anguilla Focus — 2024 government revenue projection (EC$105.53M, Premier Webster quote)
Anguilla Focus — 2025 EC$230M windfall (2025 revenue, BBC "near half of revenue" citation)
IMF — An AI-Powered Boost to Anguilla's Revenues (macro context on revenue reliance)
Domain Incite — .ai auction transition (Identity Digital takeover date, auction mechanism change, $600K figure)
DNJournal — ai.com sells for $70 million (record sale details, buyer/seller/brokers)
Axios — why HubSpot's founder sold Chat.com to OpenAI (chat.com sale price and terms)
TechCrunch — AI.com flips from ChatGPT to Elon Musk's X.ai (ai.com's earlier $11M OpenAI purchase)
AOL — Dharmesh Shah paid $700,000 for you.ai (largest .ai-extension sale at the time)
Domain Name Wire — the market for AI .com domain names (mid-tier aftermarket comps)
NamePros, summarizing NameBio data (2025 aftermarket volume, .com/ccTLD share)
The Hacker News — Phantom Squatting uses AI-hallucinated domains (Unit 42 research on hallucinated-domain exploitation)