You.com Bets Its Future on Search Infrastructure for AI Agents

From consumer search engine to the plumbing behind AI agents

You.com began as a search engine meant to compete with Google for consumers. It now presents itself mainly as infrastructure for other companies' AI. A February 2026 "Startup of the Week" profile in The Innovator called the Palo Alto company an enterprise AI productivity engine. In that account, its agents and APIs act as a base layer that businesses build their own products on, rather than a destination website for people searching.21 The company now calls itself a provider of search infrastructure for enterprise AI, and CEO Richard Socher's personal site says You.com supplies web search APIs to large language models.83

The profile is worth revisiting eight months later because events since then make the company's position clearer. Its founder has launched a second, much more expensive venture. You.com has hired a new CTO. And the case that AI agents need their own search layer is now a standard argument in the industry. In my reading, You.com's most durable asset is neither its chat interface nor its research agent. It is the search index it built itself, before most rivals saw a need for one.

What the product stack actually looks like

The Innovator describes four product lines. The first is a conversational search tool that combines several frontier models, including GPT, Claude and Gemini, with You.com's own retrieval system and returns cited answers. The second lets organizations build custom agents, and customers have reportedly made close to 100,000 of them. The third is ARI, a deep-research agent the company says reads more than 400 sources and writes a cited report in under five minutes. The fourth is a set of APIs for web search, live news, image search and private retrieval-augmented generation (RAG).21

The company's own funding announcement adds detail on the developer side. Developers can choose web and news search at different trade-offs between speed and accuracy, plus deep research, image search, private RAG and custom agent tools.11 The current site lists Contents, Answer, Research and Finance Research APIs, along with an MCP server and "Agent Skills." That shows the product is aimed at agents built on the Model Context Protocol, not just traditional software integrations.1 A recently advertised "Knowledge for Web Search" feature adds live stock prices, crypto, weather and prediction-market data at the same $5 per thousand queries.11

The main design choice is that You.com doesn't depend on any one model provider. The Innovator reports that the platform gives access to more than 40 LLMs and lets organizations send each task to whichever model fits it best.21 For enterprises, that reduces the risk of being locked into one model vendor. It also means You.com earns its value from retrieval and orchestration, not from the models themselves.

Why owning the index matters

The coverage agrees that You.com's main advantage is that it built its own search infrastructure. The Innovator says this is what separates it from competitors that rely entirely on third-party search, and quotes Socher saying an AI needs a search system that delivers the right information at the moment it's needed.21

Socher made a sharper version of this argument in a September 2025 interview. He said several rivals depend on SERP proxy networks, which route scraping traffic through consumer devices whose owners unknowingly agreed to act as IP relays. The scraped Google results are then sold as proprietary search.7 He listed three problems with that approach: large AI companies won't send queries through a competitor's systems, Google blocks the scraping, and the delays are worse than with a purpose-built index.7 He was also doubtful that the Justice Department's remedy requiring Google to share index access would help rivals soon. He expects appeals to slow it down, and he argued that Google's index was designed for ranking links, not for the dense, multi-step retrieval that language models need.7

These are claims from an interested party, and readers should treat them that way. The underlying point still holds up. If AI agents end up making more web queries than people, search built to show humans ten blue links becomes a bottleneck. The company makes exactly this case, saying today's search was designed for people clicking links, while agents need deeper context from both private and public data.11

Scale, customers and the numbers that don't quite agree

You.com raised a $100 million Series C at a $1.5 billion valuation in September 2025. Cox Enterprises led the round, and Georgian, Salesforce Ventures and Norwest also took part.16 At the time, the company said it was handling more than a billion API queries a month for customers including DuckDuckGo, Windsurf and Harvey.11 Bloomberg's reporting, as summarized by PYMNTS, also listed Databricks as a customer. It added that the money would go toward doubling a staff of roughly 100, opening a San Francisco office and growing in New York.19 A later company post goes further, saying OpenAI, Amazon and Alibaba use its API.1

The coverage is less consistent elsewhere. You.com's claim to beat the Bing and Google search APIs on accuracy and speed comes from its own internal benchmarks, as Investing.com noted.16 Headcount figures vary widely. One investor tear sheet lists 91 employees but also cites growth to 318 by November 2025.12 Revenue figures are murkier still. The same tear sheet cites a GetLatka figure of $50 million for 2024.12 CB Insights lists FY2024 revenue of about $2 million.13 That gap is too big to resolve from public data. Revenue is the least verified part of the You.com story, and any judgment of the $1.5 billion valuation should take that into account.

On the secondary market, Nasdaq Private Market estimated a share price of $6.31 as of July 20, 2026.14 In October 2025, CB Insights recorded an acquisition of Graft, a startup that had raised $19 million.13 In March 2026, You.com announced Saahil Jain as its next CTO.11

The Recursive question

The biggest change since the Innovator profile is Socher's second company. In February, the profile described Recursive as a superintelligence research lab whose plans had leaked. It said Recursive was discussing funding at a $4 billion pre-money valuation, and that Socher expected a first version of a self-improving system within one to three years.21 By May 2026, Recursive had come out of stealth with more than $650 million raised at a $4.65 billion valuation. Its investors included GV, Greycroft, Nvidia and AMD.2 So the actual round beat what was reported in February.

The Innovator stressed that the two companies are separate and independent. In its account, Socher sees AI progress as needing two tracks at once: products that sell now and long-range research.21 Even so, the two are linked in practice. Some investors appear on both lists, with GV, Greycroft and AMD all named among You.com's backers.12 Recursive's approach also depends on what You.com sells. One conference biography describes Socher's "Eureka machine" idea as coordinated agents that propose hypotheses and test them against evidence. It notes that agents can read far more material than people, which makes retrieval, freshness and provenance essential.5

The obvious risk is that the CEO's attention is split, now that Recursive is the larger company by valuation. Socher's titles across his various bios list Recursive first.910 I don't think the split is fatal, since You.com has a new CTO and a clear product focus. But it does put more weight on that leadership team to keep running the infrastructure business smoothly.

Is it a sustainable model?

People usually mean energy or climate when they call technology sustainable. For You.com, the better question is whether the business model lasts. Three things suggest it can. First, the company sells to two kinds of customer. Developers with technical skills buy the APIs directly, while organizations without in-house AI teams can buy complete systems built for them.7 Second, it says it keeps no customer data and can combine private and public data, which matters to regulated industries.16 Third, The Innovator suggests rules such as the EU AI Act could favor platforms built around citations and verifiable answers.21

The downside risks are just as real. Distribution in AI is still decided at the platform level, and big incumbents have the leverage there. Socher himself pointed to Google's reported $20 billion a year in payments to Apple to stay the default search engine, and observed that most iPhone users never change their settings.7 An infrastructure company avoids having to fight for that default spot, but it also depends on the companies that hold it.

Overall, You.com has stopped trying to be a destination and is trying to become a dependency, which is a far better position. Agent traffic is growing, its index was built before demand arrived, and its pricing is easy to understand. What it still has to show, especially with its founder also running a more expensive venture, is that its revenue matches its query volume.