Benchmark is leading an $85 million funding round in Exa Labs Inc., a San Francisco-based startup developing a search engine built for artificial intelligence agents. Bloomberg reports the deal values Exa at $700 million, a tenfold jump from its valuation in a funding round last year.

According to Bloomberg, Peter Fenton, the partner now joining Exa’s board, is one of Benchmark’s most celebrated investors. He backed Twitter when it had just 25 employees, along with other open-source players like JBoss and SpringSource. Fenton described the move as one of those rare venture moments when missing out was not an option. “The entire stack of what we call software is being reimagined as a result of the impact of AI,” Fenton said. “One of the primitives in all of software has been search.”

Benchmark has long been one of Silicon Valley’s venture firms. Founded in 1995, it pioneered an equal partnership model with no junior staff, just a small team of general partners who move fast and decide together. That tight structure helped it back eBay, Uber, Instagram, Snapchat, Twitter, Dropbox, Discord, and many more. According to Axios, in July 2024, it raised $425 million for its new fund, signaling continued confidence in early-stage bets and AI-related companies.

Exa was founded in 2021 by Will Bryk and Jeff Wang, classmates from Harvard, who set out to rethink search by and for artificial agents. The breakthrough was a model that predicts the most probable link next, not the next word, using transformer architectures trained on web link data. This means the results engine is less about keywords and SEO gaming and more about meaning and precision. This early version earned Exa $22 million in seed and Series A funding from Lightspeed, Nvidia, and Y Combinator. 

What makes Exa a different kind of search engine

Bloomberg reports that Exa’s approach is not aimed at everyday users typing queries into a search bar. Instead, it is focused on large language models and AI agents that require structured, high-quality data from across the web. Exa’s co-founder and CEO Will Bryk says traditional search, with ads and vague relevance signals, does not fit this AI-dominated world. “Traditional search, built for people typing queries, no longer makes sense in a world where the web is increasingly dominated by AI agents,” Bryk said.

This distinction also shows up in the business model. Unlike Google’s ad-driven system, which depends on clicks and engagement, Exa charges its customers per query. Bryk says this model aligns the company’s incentives with delivering accurate, useful results instead of keeping users online longer.

The company says it has built a stack that includes an in-house vector database, semantic retrieval engines, and a search API tailored to developer and enterprise use. Every query is priced, not sold with ads, so the company argues it is incentivized to deliver highly accurate and useful answers.

Customers and expansion plans

Exa already counts several major companies as customers, including Amazon Web Services, Databricks, Vercel, and Anysphere, the developer behind Cursor.

With anchor customers in place, Exa is gearing up to scale. It plans to expand the GPU cluster nicknamed “Hephaestus” to support more machine-level workloads and grow its 35-person office in San Francisco’s Mission District. Bryk wants people on the team who are excited about “building the best search in the world” during this early stage of AI’s rollout.

‍