Mercor's CEO invested in a startup three months before his own company bought it, and he says that was the plan all along.
Mercor, the AI recruiting and training-data unicorn last valued at $10 billion, announced on July 9 that it has acquired Deeptune, a San Francisco startup that builds simulated office software for AI agents to practice in. The twist: Mercor CEO Brendan Foody wrote an angel check into Deeptune's $43 million Series A back in March, a round led by Andreessen Horowitz. Three months later, his company bought the whole thing. "It was in a lot of ways the main motivation, actually," Foody told Fortune, describing the angel investment as a precursor to the deal rather than a coincidence.
Deeptune isn't a household name yet, but its pitch drew a16z partners Marco Mascorro and Martin Casado, who led that Series A. Founder and CEO Tim Lupo built what the company calls training gyms, high-fidelity reinforcement learning environments that mimic the actual software stacks knowledge workers use every day. An accountant's spreadsheet in Excel, a support rep's queue in Salesforce, a team's back-and-forth in Slack, all rebuilt as sandboxes where an AI agent can click, type, and fail without touching a real company's systems. It's closer to a flight simulator than a chatbot.
That's the piece Mercor didn't have. Mercor's network of more than five million domain experts already writes the tasks and grading rubrics that tell a frontier model whether it completed a job correctly, work that reportedly touches every Magnificent Seven company except Tesla. Deeptune builds the software those tasks actually run inside. Put the two together and Mercor now controls both ends of the pipeline that labs like OpenAI and Anthropic rely on to teach models how to use real business tools.
Neither company disclosed financial terms. Deeptune's team is relocating to Mercor's New York office, according to Fortune's reporting.
Founders write angel checks into startups all the time. They don't usually admit, on the record, that the check was step one of an acquisition plan. That's the part worth sitting with if you're an investor sizing up Mercor's next round. A CEO who personally profits from a startup his own company later buys is a governance question in any industry, and Foody didn't dodge it. He confirmed it. Whether Mercor's board or its outside investors signed off on the Series A stake before Foody wrote it isn't clear from what's been reported, and neither company has said.
The timing raises the stakes. Bloomberg reported this week that Mercor is in early talks to raise new funding at roughly a $20 billion valuation, double what it fetched in October when it closed a $350 million Series C. Foody said on X that Mercor's annualized revenue run rate hit $2 billion in June, up from $1 billion just four months earlier. A company doubling its valuation on the back of doubling revenue doesn't need to explain every acquisition to a skeptical board. Momentum like that tends to make related-party deals easier to wave through, not harder.
Mercor also isn't coming off a clean few months. In March, hackers linked to the group Lapsus$ exploited a vulnerability in the LiteLLM software supply chain and made off with an estimated four terabytes of Mercor data, including tax records, banking details, passport scans, and facial biometrics collected from contractors on its platform. That breach hasn't slowed the fundraising conversation. If anything, Mercor's revenue growth has outrun the reputational hit.
What it signals for the AI unicorn wave
Acquihiring founder-adjacent startups is becoming a familiar move among AI companies sitting on paper fortunes. Human-data rivals like Scale AI and Surge AI have made smaller acquisitions of their own to shore up infrastructure ahead of funding talks. Mercor's version is more direct: buy the team building the environments your product depends on, and move fast enough that you can point to a finished integration the moment investors ask what the next round actually buys.
For founders competing in the same lane, the message is blunt. The market for AI-agent training infrastructure is consolidating before most people have even heard of the companies doing the consolidating. Deeptune went from closing a Series A to becoming a subsidiary in under four months, faster than most startups take to hire a VP of sales.
Also read: Meta Opens Muse Spark 1.1 to Developers While Locking Rivals Out at Home • China finally lets its AI giants buy Nvidia's H200 chip on its own terms • Mercor's Revenue Hit $2 Billion in June and the AI Data Race Isn't Slowing
