- Exa announced a $250M Series C at a $2.2B valuation on May 20, the morning after Google declared its own search box obsolete at I/O 2026.
- The raise is a direct declaration that AI agents have already overtaken humans as the web’s primary information consumers, and that the entire existing search stack was built for the wrong user.
- Exa’s crawlers index over 500 billion URLs using models it built from scratch, unlike most competitors, which are wrappers reselling Google or Bing results.
On the morning of May 20, Will Bryk published a blog post announcing that his company Exa had just raised $250M at a $2.2B valuation. The previous evening, Google had stood on stage at I/O 2026 and declared the search box it invented obsolete, replacing it with AI agents that monitor the web around the clock, synthesise answers in real time, and never wait for a human to type a query. AI Mode has crossed one billion monthly users. Queries are more than doubling every quarter. The search box, Google confirmed, is done.
Bryk’s opening line that morning: “AI agents will search the web more than humans this year.”
Whether the timing was coordinated, opportunistic, or simply inevitable depends on your read of the moment. What is not debatable is that two of the most consequential announcements in the history of search landed within twelve hours of each other and pointed in exactly the same direction.
The problem nobody named until agents arrived
For twenty-five years, search had one implicit assumption baked into its architecture: a human was on the other end. Google’s PageRank model ranked pages by how often other pages linked to them, a proxy for human editorial judgement at web scale. It worked spectacularly, until the human disappeared from the loop.
AI agents do not click links. They do not browse. They do not tolerate SEO spam or ad-optimised summaries. They need accurate, contextually rich information retrieved at machine speed, for queries that run to paragraphs rather than keywords. The latency requirements, the volume, and the tolerance for noise are all categorically different to consumer search. Most search API providers available to developers today are wrappers — they query Google or Bing and reformat the results. That architecture has a hard ceiling. Quality, latency, and cost are determined entirely by the upstream provider, which built for human scale, not agent scale.
Bryk and co-founder Jeffrey Wang met as Harvard freshmen. They built their first search engine as roommates, bought $1M worth of GPUs before the shortage, and trained a model not to predict the next word — the standard transformer approach — but to predict the next link, learning from how humans share information across the web. The result retrieves semantically rather than by keyword, comprehensively rather than by popularity, and fast, sub-200 milliseconds for real-time agents. “Six months ago we were worse than Google at code search,” Bryk wrote on May 20. “Now we’re used by nearly every coding agent.”
The competitive picture, reshuffled
Three things happened in the months before May 20 that reshaped the market Exa is entering. Tavily, purpose-built for agents, over one million developers, was acquired by AI cloud company Nebius in February 2026 for up to $400M, turning an independent infrastructure vendor into a proprietary component of a larger platform. Developers who do not want their search layer tied to a cloud provider now have one fewer independent option. Perplexity, valued at $20B after a $500M Series E in May 2025, occupies a different lane: its Sonar API bundles LLM synthesis with retrieval, which adds cost and reduces control for developers who already have a reasoning model and need only the underlying data. Then Google I/O happened, and confirmed that the category Exa is building for is not hypothetical.
Google’s announcement that AI Mode queries are doubling every quarter, and that persistent background agents will soon search the web continuously on behalf of users, is the clearest market sizing signal the agentic search industry has ever received. The person who stood on stage and said search volume just hit an all-time high despite AI Overviews accidentally described Exa’s total addressable market. The AI search engine market sits at $18.5B today and is projected to reach $66.2B by 2035, a 14% CAGR according to Future Market Insights. The agentic segment specifically is growing at a 43.8% CAGR through 2034, according to Precedence Research.
What the $250M pays for
Andreessen Horowitz, which previously backed Airbnb, GitHub, Stripe, and OpenAI led the round, with returning investors Benchmark, Lightspeed, Nvidia’s NVentures, and Y Combinator also participating. The a16z team’s investment memo was pointed: “The first search wars were won by organising information for people. The next will be won by organising information for agents.” The capital goes toward training next-generation retrieval models, scaling infrastructure to handle hundreds of thousands of searches per second, and a global go-to-market push led by incoming CRO Marcus Holm, former president of LaunchDarkly. Exa currently serves over 5,000 companies and 400,000 developers, including Cursor, Cognition, HubSpot, and Monday.com. It has grown from a $700M valuation at its $85M Series B in September 2025 to $2.2B in eight months.
Bryk’s framing of the stakes goes well beyond infrastructure. “Information is critical civilisational infrastructure for our new AI reality,” he wrote. “Politics is fragmenting, wars are raging, and AI is accelerating, we need tools we can trust. If we can build perfect search so that every AI has the highest quality information, then every human will too.”
That is either the most important infrastructure argument in technology right now, or a very expensive philosophy project. Google just bet that search belongs inside its own ecosystem, absorbed into agents it builds and controls. Exa just raised $250M the morning after betting that independence is worth paying for, and that the question of who controls what AI agents know about the world is too important to leave to the company that already controls what humans do.
