AI Tools Statistics 2026: What Cataloguing 8,500+ Tools Taught Us

Most AI tools statistics floating around the web are recycled estimates of estimates. We have something better: our own database. PopularAiTools.ai catalogues 8,900+ AI tools and resources — 327 products we've reviewed in depth, plus 8,583 ecosystem items (MCP servers, skills, agents, CLIs) we track continuously. In July 2026 we ran the analysis across all of it. Some of what we found surprised us; the domain-name result genuinely made us re-run the query.

The dataset (and its limits)

Two populations, analysed separately. First, 327 curated tool reviews — products we've individually tested and written up, spanning 20+ categories; 318 of them have live product URLs we could analyse. Second, 8,583 ecosystem items — the MCP servers, Claude Code skills, agents, commands, hooks, and CLIs our crawlers catalogue. The obvious caveat: our directory reflects what we choose to review and what the Claude-centric ecosystem publishes, not a census of every AI product on Earth. It's a large, continuously-maintained sample — treat percentages as directional for the wider market.

.ai has overtaken .com — the quiet domain flip

Of 318 tools with product URLs, 123 (38.7%) live on a .ai domain versus 118 (37.1%) on .com. That's a five-tool margin — statistically a coin flip — but the direction of travel is unmistakable: .com held a comfortable lead in every informal count we did through 2024. Anguilla's national TLD is now the default address of the AI industry, .io (6.9%) has fallen to a distant third, and .dev (4.1%) quietly became the choice of the infrastructure crowd. If you're naming an AI product in 2026, the data says your customers won't blink at a .ai — and the .com is probably parked by a squatter anyway.

The MCP monoculture: 83 servers for every agent

This is the stat we'd put on a conference slide: of 8,583 ecosystem items we track, 6,704 — 78.1% — are MCP servers. Skills are a distant second at 1,424 (16.6%), and packaged agents barely register at 81 (0.9%). That's 83 MCP servers for every published agent.

The interpretation matters: the ecosystem's energy isn't going into shipping finished agents — it's going into connectivity. Everyone is wiring AI into databases, SaaS APIs, browsers, and file systems, and letting general-purpose agents (Claude Code chief among them) do the orchestration. If you're deciding what to build, the whitespace is obvious: polished, opinionated agents and skills are massively underrepresented relative to plumbing. Our MCP servers directory and skills directory hold the underlying lists.

What actually gets built: coding first, creative in aggregate

Coding is the single biggest category at 13.8%, with automation right behind at 11.0% — together, a quarter of everything we review is developers building for developers. But group the creative categories (image 8.0%, video 7.6%, voice 7.3%, design 4.0%, music 1.5%, 3D 1.8%) and they total 30.3% — the creative stack in aggregate out-builds the dev stack. The underweight surprise: education (1.5%) and finance (0.3%), two enormous industries with strikingly thin AI tooling relative to their size. Founders looking for uncontested ground, take note.

Naming: the "AI" prefix refuses to die

20.8% of tools — one in five — still include "AI" in their product name. The conventional branding wisdom says this is dated (nobody named their company "InternetBooks.com" after 1999, except the ones who did). The data says founders don't care: the proportion has barely moved in our directory over the past year. Combined with the .ai domain result, the picture is an industry still loudly labelling itself. We'll re-run this annually; when the "AI"-in-name share starts falling, that's the maturity signal.

FAQ + citing this data