On May 20 at 4:44 p.m. Pacific time, San Francisco-based Exa Labs completed a $250 million Series C round at a $2.2 billion post-money valuation, led by a16z.
The numbers alone are striking: the company completed its Series B last fall at a $700 million valuation, meaning it has more than tripled in six months.
A product shaped by 5,000 companies
Exa does one narrow thing: it provides a search API for AI applications. Not a search box for humans, but an interface for agents. When you build an AI agent to find "all review articles about GPT-5.5 from the past week," it calls an API like Exa's behind the scenes.
Its customer list is worth noting: Cursor, Cognition, HubSpot, OpenRouter, Monday.com, and more than 5,000 other companies. Among them, Cursor and Cognition are among the highest-valued AI coding companies today, and they outsource search to Exa. This kind of "upstream infrastructure" signal is more telling than the funding itself. Over 400,000 developers use Exa's platform.
CEO's key quote
Will Bryk said in an interview:
"As trillions of agents come online over the coming years, search needs will grow thousands of times beyond total Google search volume."
He added:
"Most other search providers actually wrap other search engines and therefore cannot compete on quality/latency/cost."
In other words, most AI search providers essentially wrap Bing or Google APIs, so they can't match the quality, latency, or cost of a self-built engine.
What Exa does technically
| Product | Function | Key Metric |
|---|---|---|
| Exa Instant | Speed-optimized search | 180ms per query |
| Contents | Fetch full webpage text | Provides context for agents |
| Exa Agent | Multi-step search tasks | Self-decomposes and follows up |
Under the hood is a proprietary vector database that can search billions of embeddings in a tenth of a second. The embedding model runs on Exa's own NVIDIA GPU cluster. In short, from crawling, indexing, vectorization to query, every layer is Exa's own—not a shell.
What Google is doing
The timing is interesting. The day before Exa's round, Google revamped Search at I/O, calling it the "biggest change in 25 years." The new version accepts text, images, video, and Chrome tabs as input, and outputs synthesized answers instead of link lists. But Bryk's assessment is blunt: that's a redesign for humans, not for AI.
Humans click a few times per search. Agents may send 50 search requests per task, each requiring structured, parseable data that can feed directly into the next prompt. The two demand curves are fundamentally different.
Where the money goes
Bryk stated clearly in the announcement: expand AI infrastructure, train next-generation models, and scale the system to handle hundreds of thousands of searches per second. The rest goes to hiring, with a focus on go-to-market—meaning a sales battle in the coming year.
So this isn't just another AI company raising big money. It's about search, the internet's oldest gateway, splitting into two product lines: one for humans, one for agents. Google defends the first; Exa is racing for the second. Whether the $2.2 billion valuation is justified—we'll see next year.
Sources: Exa Labs raises $250M at $2.2B valuation for its AI search tools (SiliconANGLE); CocoLoop; Exa Raises $250 Million for AI-Powered Search Infrastructure (PYMNTS); Andreessen-Backed AI Search Startup Exa Valued at $2.2 Billion (Bloomberg)