When Will Bryk told his father in 2021 that he wanted to build a search engine, the response was incredulous. “What’s wrong with you? Have you heard of Google?” Bryk recalled in a recent interview. Yet four years later, the company he co-founded, Exa Labs, has raised an $85 million Series B round led by Benchmark at a $700 million valuation.
The financing marks a milestone for Exa, which has positioned itself as a new kind of search provider designed not for humans but for artificial intelligence systems that increasingly power applications across industries.
From Research Project to Infrastructure
Bryk, who studied physics and computer science at Harvard, started Exa with classmate Jeff Wang. The pair bought their own GPU cluster and spent more than a year experimenting with algorithms for web search. Their bet was that keyword-based indexing, the foundation of Google, would not be sufficient for AI agents that need to parse, structure, and process vast amounts of online information.
“We are built for this world, that is the key thing. We are built for AI from scratch,” Bryk said.
Exa’s core product is an application programming interface that developers can integrate directly into their systems. It offers two modes: a fast, low-latency option and a more compute-intensive tool called Websets, capable of assembling large, structured datasets from the web. Both are designed to feed large language models with real-time, high-quality data. The company says its software can process more than 100 queries per second with sub-450 millisecond latency.
Unlike consumer-facing search engines, Exa charges per query and maintains a strict “zero data retention” policy, a feature aimed at enterprise users handling sensitive information. Customers include Databricks, Amazon Web Services, Vercel, and AI-native companies such as Cursor.
Benchmark’s Stretch
Benchmark’s involvement underscores how central investors believe AI-native search could become. The venture firm is known for sticking to early-stage deals of around $15 million, yet Benchmark partner Peter Fenton committed $50 million personally to Exa’s Series B.
“You have those moments in venture where you feel like, ‘Oh boy, now I know I have to invest. I better not lose the opportunity,’” Fenton said.
Fenton will join Exa’s board, bringing with him a track record that includes early bets on Twitter and Yelp. He described Exa as addressing one of the “primitives in all of software,” a category shift that incumbents may not be able to navigate.
The round also included Nvidia, Lightspeed, and Y Combinator, each of which has strong ties to the AI ecosystem. Nvidia’s participation is particularly notable, as Exa runs its infrastructure on a large cluster of the company’s H200 GPUs. Exa plans to expand that cluster fivefold with the new funding.
Scaling Toward “Perfect Search”
Bryk’s ambition goes beyond building a competitive developer tool. “I wanted to organize all the world’s information for real,” he said. With the Series B, he frames Exa as entering “stage two,” which involves scaling infrastructure to cover more of the web and refining methods for natural-language retrieval.
The task is formidable: storing and indexing hundreds of billions of web pages while filtering out spam, misinformation, and low-value data. Exa believes its clean business model, no advertising, usage-based pricing, aligns its incentives with information quality. That positioning has attracted early adoption in industries where precision and privacy matter, including private equity and consulting.
Competitively, Exa is entering a crowded field. Google continues to experiment with AI-powered search, and startups like Perplexity are racing to build consumer-facing alternatives. Exa’s differentiation lies in its focus on AI agents rather than end users, making it more of a behind-the-scenes infrastructure provider than a household brand.
For now, Exa is a 35-person company operating out of San Francisco’s Mission District, where Bryk has outfitted the office with nap pods to accommodate the pace of work.
