Exa raises $250M at $2.2B valuation to build search engine for AI agents, not humans

Exa Secures $250M Series C Funding to Scale AI Search Infrastructure

Exa has closed a $250 million Series C funding round led by Andreessen Horowitz, reaching a valuation of $2.2 billion2

2

. The investment comes less than a year after the company's $85 million Series B round, which included contributions from Nvidia and Y Combinator2

2

. This rapid fundraising trajectory signals growing demand for AI-powered search infrastructure as intelligent agents become the primary consumers of web information.

The funding will enable Exa to expand its AI infrastructure, train next-generation models, and scale systems to support hundreds of thousands of searches per second2

2

3

3

. The company also plans to hire more go-to-market professionals to support its expanding customer base2

2

.

Building a Search Engine for AI Agents, Not Humans

Unlike traditional search engines designed for human users, Exa provides a web search API specifically optimized for AI agents3

3

. The company operates its own independent search engine rather than acting as a wrapper for existing providers like Google Search3

3

. This distinction matters because AI agents have fundamentally different search requirements than humans.

Will Bryk, Exa's co-founder and CEO, explains the scale challenge: "As trillions of agents come online over the coming years, search needs will grow thousands of times beyond the total search volume of Google"3

3

. Agents don't ask simple queries. They probe long-tail, constantly shifting information with complex queries that can span paragraphs1

1

. As one leading AI company noted, "In the limit, if we could search 100% of the time, we probably would. It just comes down to GPUs, latency, and cost"1

1

.

Technical Architecture Behind AI-Powered Search Tools

Exa's fastest offering, Exa Instant, completes queries in under 180 milliseconds, making it the speediest search service of its kind on the market according to the company2

2

. This low-latency performance matters acutely in user-facing agent flows, especially for time to first token1

1

.

The company stores web data in a custom vector database that can query billions of embeddings in one tenth of a second while using less memory than a high-end personal computer2

2

. One contributor to the database's speed is storing certain important files in CPU cache instead of RAM2

2

.

Exa ingests public web data using a custom software platform called exa-d, which parallelizes key data processing tasks across multiple graphics cards2

2

. The platform skips many unnecessary file changes that data management systems often make when updating records, further optimizing hardware utilization2

2

. The company trains custom neural networks called embedding models on an in-house cluster of Nvidia graphics cards to turn web data into mathematical structures that AI models can understand2

2

.

Multi-Step Search Workflows and Enterprise Adoption

Beyond Exa Instant, the company offers several other AI-powered search tools. An offering called Contents enables AI applications to retrieve the full text of webpages2

2

. Exa Agent facilitates multi-step search workflows, allowing market research agents to find an e-commerce store's most popular products and then enrich the list with customer feedback from reviews sites2

2

.

Since launching its AI-focused API in early 2023, Exa's customer base has grown to more than 5,000 companies and over 400,000 developers2

2

3

3

. These users include leading AI companies from startups on the frontier like Cursor and Cognition to large enterprises like HubSpot, Monday.com, OpenRouter, and many Fortune 500 companies1

1

3

3

.

Andreessen Horowitz noted that developers and AI agents are reaching for Exa first, with one customer stating, "this is the default for getting agents to do web search now"1

1

. Bryk added that six months ago Exa was worse than Google at code search, but now it's used by nearly every coding agent3

3

.

The Vision Behind Building Perfect Search

Will Bryk and co-founder Jeffrey Wang built a search engine in their dorm at Harvard a decade ago1

1

. Years before ChatGPT or the AI boom really started, they were inspired by the transformer breakthrough and believed deeply that AI would fundamentally change how we access information1

1

. They set out to build the search engine for a future where agents become the primary consumers of the web1

1

.

Bryk articulates the company's mission: "We're organizing the world's knowledge, but this time for AI"1

1

. As agents make increasingly important business decisions, their requirements for comprehensiveness, freshness, and precision will far exceed what humans require3

3

. Agents will need perfect search over all the world's information at an unprecedented scale3

3

.

Andreessen Horowitz believes the first search wars were won by organizing information for people, but the next will be won by organizing information for agents1

1

. As AI both consumes and produces the internet, the web will become bigger, faster, and noisier, making search the compass for agents to navigate it1

1

.