Exa funding: Every round, investor, and valuation (2026)

Exa builds a search engine that AI models use instead of people. Developers call its API when a chatbot, coding assistant, or research agent needs fresh information from the web, and Exa returns full pages of relevant content rather than a list of blue links and ads. Customers pay per request.

Exa has raised $357 million across four rounds since it started in 2021 as Metaphor, all of it equity. Its latest, a $250 million Series C led by Andreessen Horowitz in May 2026, valued the company at $2.2 billion.

The company's pitch is that it runs its own index. Most search APIs resell results from Google or Bing, while Exa crawls the web, embeds the pages, and ranks them with models it trains on its own GPU cluster. Let's dive into Exa's funding history.

Exa spent its first three years on a small seed round and its next two raising more than 70 times that amount. The jump came once AI agents gave its product a large market of paying developers.

Will Bryk and Jeff Wang, college roommates at Harvard, started the company in 2021 under the name Metaphor. Bryk had been an early engineer at Cresta, a customer service AI company, and Wang had built data infrastructure at Plaid. They went through Y Combinator's Summer 2021 batch.

The seed round raised $5 million. Exa has confirmed the amount, but trackers disagree on the date, putting it anywhere from September 2021 to January 2022, and Y Combinator is the only investor every source names.

The idea was to search the web by meaning rather than by keyword, using the same transformer models behind large language models. Metaphor launched its first search engine in November 2022, the same month ChatGPT came out. That timing turned a consumer search experiment into a tool for the developers suddenly building on language models.

In January 2024, Metaphor renamed itself Exa. The name refers to 10 to the power of 18, a deliberately smaller number than the googol, 10 to the power of 100, that gave Google its name. The company then had about seven employees.

Lightspeed Venture Partners led the $17 million Series A in July 2024, with Nvidia's venture arm, NVentures, and Y Combinator joining. Exa said revenue had tripled in the few months before the round. Databricks was among its named customers, using Exa to gather training data for its models. That fit a company whose Databricks funding rounds were paying for a push into AI products.

Benchmark led the $85 million Series B in September 2025, and partner Peter Fenton joined the board. Lightspeed, NVentures, and Y Combinator all returned.

The round followed a product push. Exa launched Websets in early 2025, a tool that builds lists such as every startup that fits a set of criteria, and Exa Research, an API that runs multi-step research jobs. A month before the round, Microsoft retired its Bing Search API, which had been one of the main sources of web results for developers outside Google.

The Series B money went into three places: a larger index, a bigger GPU cluster, and more hires across engineering and sales. Three weeks after the round, Exa released exa-code, a free, open-source tool that lets coding agents search for working code examples, aimed squarely at its fastest-growing customer group.

Andreessen Horowitz led the $250 million Series C in May 2026, eight months after the Series B. Benchmark, Lightspeed, and Y Combinator joined again, and the a16z team on the deal included general partner Sarah Wang.

By then, Exa said more than 5,000 companies and 400,000 developers used its API, with customers including Cursor, Cognition, HubSpot, OpenRouter, and monday.com. Its crawlers were tracking more than 500 billion URLs. Exa has not raised again since, and no debt, secondary sale, or tender offer has been reported.

Exa's valuation rose roughly 30-fold in under two years, measured from the implied Series A price. Most of that came in a single step, when Benchmark priced the Series B.

Exa never disclosed a Series A valuation. Bloomberg reported that the Series B came at ten times the price of the previous round, which implies about $70 million. The Series C then priced the company at about 3.1 times the Series B. At $2.2 billion, Exa was valued at a little over six times the capital it had raised.

The gap between the two later steps reflects what investors were paying for. At the Series B, Exa was a small company with an unusual technical bet. By the Series C, AI agents had become the main way many developers used language models, and those agents needed a search engine to call.

Exa does not publish revenue. Research firm Sacra estimated annualized revenue of about $10 million in September 2025, roughly 11 times the year before, which would put the Series B at about 70 times revenue. That is high even by the standards of a market where AI startups reach unicorn status in months, and it shows how much of the price rests on future search volume.

The cap table is less public than the valuation. No investor has disclosed its stake, and Exa has not said whether any of its rounds came with special terms.

Exa's investors fall into three groups: the accelerator that backed it first, the venture firms that led each priced round, and Nvidia, which sells the chips Exa trains on.

Y Combinator backed Metaphor in its Summer 2021 batch and has put money into every round since. It is the only investor on the cap table from the start, through the years when the company was a small search experiment with no clear market. Its standard deal at the time paid $125,000 for a fixed slice of each company in the batch.

Lightspeed led the Series A in 2024 and returned for both later rounds. The firm backed Exa when its customers were mainly AI developers and labs buying training data, before AI agents became a mainstream product category.

Benchmark led the Series B, and Peter Fenton took a board seat. Fenton is known for early bets on developer infrastructure, and Bloomberg framed the deal as a wager that Exa could become the Google of the AI era.

Benchmark also returned in the Series C, which kept the firm's stake from shrinking as the valuation tripled. Fenton is the only investor board member named in any public announcement.

Andreessen Horowitz led the Series C, with Sarah Wang, Jennifer Li, Stephenie Zhang, and Jason Cui writing the firm's investment note. The firm's thesis is that agents will need a search layer of their own, and that a provider with its own index can win that market.

That thesis treats search as one layer of a wider agent stack. Frameworks such as LangChain, whose agent framework funding followed a similar path, help developers build agents, and Exa supplies the web data those agents look up.

NVentures, Nvidia's venture arm, joined the Series A and the Series B. It is not named in the Series C announcements. For Nvidia, Exa is both a portfolio company and a customer: its GPU cluster runs on Nvidia H200 chips, the hardware behind Nvidia's data center business.

Exa spends most of its capital on three things: the GPUs that train its models, the crawling and indexing systems that hold its copy of the web, and the engineers who build both.

In May 2025, Exa switched on the Exacluster, a $5 million cluster of 144 Nvidia H200 GPUs. The company said training runs that used to take about a week now take about a day. The cluster embeds billions of web pages and trains the models that rank them.

The Series B was earmarked in part to grow that cluster fivefold. The Series C goes into training the next generation of Exa's search models. Exa also runs on Amazon Web Services for hosting, so it pays for cloud capacity on top of its own hardware.

Running its own index is Exa's main cost and its main selling point. It writes its own crawler, its own vector database, and a data framework called exa-d, which it described in January 2026.

The scale is growing fast. Exa reported tracking more than 500 billion URLs at the Series C. In August 2026, Bryk said Exa now serves 80 billion pages and aims to reach Google's scale by early 2027, according to Runtimewire. The Series C also pays to scale the system to hundreds of thousands of searches per second.

Exa had about seven employees in early 2024 and is now estimated at somewhere between 75 and 120, depending on the source. The Series C funded senior hires from search incumbents, including a head of retrieval infrastructure from Meta, a head of search backend from Yandex, and a research team from Google.

The company is also building a sales team for the first time. Marcus Holm, former president of software company LaunchDarkly, joined as chief revenue officer with the Series C. In March 2026, Exa opened its first Asian office, an engineering hub in Singapore.

Exa makes money by charging per request. Developers get $10 of free credits a month, and a standard search costs $7 per 1,000 requests, with deeper research searches priced higher. Large customers negotiate custom contracts.

That model means every new agent product that calls Exa adds revenue without a new sales cycle. It also means Exa's margins depend on how cheaply it can run each search, which is why so much of the money goes into its own hardware and index rather than paying another search engine for results.

Exa's growth focus is enterprise. Coding agents such as Cursor and Cognition are its best-known customers, and the new sales team targets larger contracts with companies such as HubSpot, private equity firms, and consulting firms. In April 2026, Google Cloud added "Grounding with Exa Web Search" to its Gemini Enterprise platform in private preview, which puts Exa in front of corporate buyers that would otherwise default to Google's own search business.

The product line keeps widening from a single search API into deeper agent tools. Exa has added an academic paper index, a free code search tool for coding agents, Monitors that track pages for changes, and Exa Agent Ultra, a top-tier research product launched in September 2026. Competition is moving just as fast: Parallel Web Systems, founded by former Twitter CEO Parag Agrawal, raised at a $2 billion valuation in April 2026, Nebius agreed to buy Tavily, and Perplexity's funding history gives it the cash to push its own Sonar search API. Brave, which runs an independent index too, sells search results to developers as well, under Brave's founder-led ownership.

Exa has no public IPO plans. The main risks are that Google and the large model labs build search directly into their models, that price competition squeezes a market of usage-based contracts, and that customers such as Cursor's fast-growing coding business move to another provider. The next round will likely depend on whether the Series C spending gets Exa to the index size it has promised and turns more developers into large enterprise accounts.

Exa has raised $357 million across four rounds, all of it equity. That includes a $5 million seed, a $17 million Series A in July 2024, an $85 million Series B in September 2025, and a $250 million Series C in May 2026. The company has not reported any debt.

Exa's lead investors are Andreessen Horowitz, Benchmark, and Lightspeed Venture Partners. Other backers include Y Combinator, which has joined every round, and NVentures, Nvidia's venture arm, which invested in the Series A and Series B.

Exa was valued at $2.2 billion after its May 2026 Series C. That is up from $700 million at the Series B in September 2025 and about $70 million at the Series A in 2024, according to Bloomberg. The company has not disclosed a valuation since.

No. Exa is private and has not filed for an IPO. Its shares are not traded on any stock exchange. The legal entity is Exa Labs, Inc., based in San Francisco.

Exa is owned by its founders, employees, and investors. Co-founder Will Bryk is CEO, and Jeff Wang is the other co-founder. The largest outside holders are likely the firms that led its rounds, Andreessen Horowitz, Benchmark, and Lightspeed, along with Y Combinator, though exact stakes are not public.