Exa Labs

Exa Labs Inc. was founded in San Francisco in 2021 as Metaphor Systems by Harvard classmates and close friends William (Will) Bryk and Jeffrey (Jeff) Wang. Bryk studied computer science and physics, did machine-learning research, led Harvard's robotics club and worked as a software engineer at conversational-AI company Cresta; Wang studied computer science and philosophy and worked at Plaid from 2019 to 2022. They entered Y Combinator's Summer 2021 batch with the belief that transformer-based semantic retrieval could produce a better search engine than keyword/ad-optimized Google. The founders bought and assembled a GPU cluster, built crawlers, an index and vector-database infrastructure, and launched their first search product in November 2022—two weeks before ChatGPT. Requests from early AI-app builders for API access changed the primary market from a human search experience to search infrastructure for LLMs and agents. Metaphor rebranded as Exa in early 2024. Exa's central claim is that AI systems require full page content, structured data, lower latency, higher recall and controllable domain/time/category filters rather than a consumer results page optimized for clicks and advertising. It crawls and indexes the web itself instead of simply wrapping Google/Bing, trains specialized embedding/retrieval/highlight models and offers different compute/latency tradeoffs. By May 2026 Exa said its crawlers tracked more than 500bn URLs, more than 5,000 companies and 400,000 developers used the service, and customer logos included Cursor, Cognition, HubSpot, OpenRouter, Monday.com and Sarvam; customers also include private-equity, consulting and Fortune 500 organizations, but those broader claims are not customer-level audited disclosures. Current products include Search API modes from Instant/Fast/Auto to Deep/Deep Reasoning; Contents API for crawling, full text, highlights and structured extraction; Exa Agent for multi-step cited web research and structured outputs; Websets/entity indexes for comprehensive people/company/list building and enrichment; Monitors for recurring change detection; an Answer endpoint; MCP/connectors and more than 50 integrations; and Exa Connect for external data providers. Google Cloud launched Grounding with Exa in private preview and listed the Exa Agent in Gemini Enterprise's Agent Marketplace in April 2026. Search infrastructure is usage priced: current Search with contents is $7 per 1,000 requests, Contents starts at $1 per 1,000 pages per content type, Deep costs $12 per 1,000 and Deep Reasoning $15 per 1,000. Agent effort tiers run from $0.012 minimal to $1.00 xhigh per request. New users receive $20 initial credits and $10 monthly; enterprise contracts add custom QPS/concurrency, up to 1,000 results, SSO, SLA, support, zero-data-retention and HIPAA arrangements. This aligns incentives with retrieval quality rather than ads, while also making gross margin sensitive to crawling, model inference and egress. Funding is four economic rounds totaling $357m on announced figures: an approximately $5m seed including YC and Pioneer-era backers; a $17m Series A announced July 16, 2024 led by Lightspeed's Guru Chahal with NVentures and YC, taking cumulative funding to $22m; an $85m Series B on September 3, 2025 led by Benchmark at a $700m post-money valuation with Lightspeed, YC and NVentures, after which Peter Fenton joined the board; and a $250m Series C on May 20, 2026 led by Andreessen Horowitz at a $2.2bn post-money valuation, with Benchmark, Lightspeed, YC and angels reported. Sarah Wang/a16z reportedly joined the board. SEC/secondary share-class data show closings split across preferred subseries and amounts around $352.96m actually sold, which should not be mistaken for additional rounds; announced headline total is used because the 2024 and 2026 company rounds are defined that way. The legacy database's $101.95m captured only two Form D floors and missed the entire Series C. Capital will expand next-generation search models, infrastructure for hundreds of thousands of searches per second and global go-to-market. The company reported 75 people on YC's current profile and around 100-120 in Series C-era secondary coverage; the reliable bounded range is 51-200. Marcus Holm, formerly president of LaunchDarkly, joined as chief revenue officer in May 2026. No audited revenue, ARR, margin, burn or profitability has been disclosed. A secondary estimate of roughly $10m ARR in September 2025 is plausible but not primary enough to store as fact. Competitive providers include Google/Bing search APIs, Brave Search, Perplexity, Tavily, You.com, Parallel, SerpAPI, Diffbot, Firecrawl, Apify and internal crawling/RAG stacks. Exa differentiates through a proprietary web-scale index, semantic/neural search, purpose-built code/people/company/scholarly indexes, latency modes, full content/highlights and enterprise ZDR rather than consumer answers/ads. Its benchmarks show strong accuracy/latency and token reductions, but they are company-selected evaluations; definitions, query sets, caching, freshness and competitors' configuration can materially affect results. Building an independent search engine is capital- and operations-intensive: crawling hundreds of billions of URLs requires continuous compute/storage/network spend and defenses against spam, poisoning, malicious pages and adversarial SEO. Results can be stale, incomplete, biased, unsafe or wrong; full-page extraction and research synthesis can reproduce copyrighted or paywalled material, hallucinate attribution or reveal personal data. Web scraping law, publisher robots/opt-out rules, contract, trespass, copyright/database rights and privacy regimes are evolving globally. Exa says services often return public data and offers privacy rights, SOC 2 Type II and optional ZDR/HIPAA controls, but public people/company indexes and Websets create GDPR/CCPA accuracy, deletion, purpose-limitation and profiling exposure. Query logs can reveal confidential strategies unless ZDR is contracted. Exa was not named in the 2024 TollBit allegations about specified AI scrapers and no material Exa-specific lawsuit, regulator action or confirmed breach was found in the manual review; industry litigation involving Perplexity/SerpAPI and publishers is a forward risk, not evidence of Exa misconduct. Dependence on GPU/cloud capacity and partnerships, rapidly improving incumbents, customer concentration among AI coding vendors, price compression and a $2.2bn valuation far ahead of disclosed financials are material commercial risks. Its mission language—'perfect search,' agents searching 1,000x Google and search needs reaching extraordinary scale—is strategic aspiration, not a forecast to treat as a metric. Key diligence gaps are actual revenue/ARR and margins, index coverage/freshness by region/language, crawl compliance and licensing policy, customer concentration/retention, benchmark reproducibility, board/cap table and the exact sold amount/syndicate of the Series C.

Founding story

Harvard friends Bryk and Wang believed search quality and information access would shape civilization, entered YC S21 and spent years building a transformer-era search engine before demand existed. ChatGPT launched two weeks after their first public engine and redirected them toward an API for AI agents.

Business model

Developer and enterprise AI infrastructure: proprietary web search/indexing, content extraction, structured entity data, monitoring and research agents delivered through usage-based APIs and enterprise contracts.

Pay-as-you-go per search/page/agent run plus enterprise minimums/custom commitments for QPS, concurrency, result volume, ZDR, SSO, HIPAA, support and SLAs; no ads or consumer subscription dependence.

Traction

Company reports 5,000+ companies and 400,000+ developers by May 2026, 500bn+ URLs tracked, and production use by Cursor, Cognition, HubSpot, OpenRouter and Monday.com. Team about 75 on YC, with later estimates near 100-120. Benchmarks and 20x/94% token-efficiency claims are first-party and methodology-dependent; financial scale is undisclosed.

Latest developments

After a $250m Series C at $2.2bn in May 2026, Exa launched its Agent API in June and a scholarly-publication search update in July; it is scaling infrastructure toward hundreds of thousands of searches per second and global GTM. Google Cloud added Exa grounding and Gemini Enterprise marketplace distribution in April.