Mercor, the AI talent and data platform valued at $10 billion, has acquired Deeptune, a startup that builds simulation environments where AI agents learn to use enterprise software before being deployed in the real world. The deal, announced on July 9, was not exactly an arm’s-length transaction. Mercor CEO Brendan Foody had quietly helped fund Deeptune months earlier.

Financial terms weren’t disclosed. But the strategic logic is clear: Mercor wants to own the full stack of AI training infrastructure, from the human experts who label data to the virtual sandboxes where AI agents practice clicking through Salesforce.

What Mercor is actually building

Mercor started life in 2023 as a programmer-matching platform. That pivot turned out to be spectacularly well-timed. The company went from roughly $1 million in revenue to hundreds of millions, driven by insatiable demand from AI labs hungry for high-quality human feedback. By October 2025, Mercor had raised a $350 million Series C that pegged its valuation at $10 billion. The company reportedly spends over $1.5 million per day on expert training work for AI models.

Deeptune fills a specific gap in that pipeline. The Andreessen Horowitz-backed startup builds simulation environments where agents can practice real-world tasks like navigating Excel spreadsheets, managing Slack workflows, and handling CRM operations.

Deeptune raised a $43 million Series A in March 2026, led by Andreessen Horowitz. Foody participated in that round as an angel investor. Months later, he acquired the whole company. The Deeptune team will relocate to New York as part of the deal.

This wasn’t Mercor’s first acquisition spree, either. The company previously acquired Sepal AI in February 2026, suggesting a deliberate strategy of rolling up specialized AI training tools under one roof.

What investors should watch

Foody’s trajectory is worth noting on its own. He founded Mercor at 19, and at 23 he’s running a $10 billion company that’s burning through seven figures daily on AI training operations.

The fact that Foody invested in Deeptune before acquiring it raises governance questions that sophisticated investors will want to examine. Insider-funded-then-acquired deals aren’t inherently problematic, but they do create information asymmetries that matter when valuations are this large and terms are undisclosed.

For the broader AI infrastructure market, the acquisition confirms that simulation environments for agentic AI are becoming a critical bottleneck. As AI systems move beyond text generation into actually performing tasks inside enterprise software, the training methodology has to evolve. You can’t teach an AI to navigate a complex CRM by showing it screenshots. It needs to interact with realistic environments, make mistakes, and learn from them, all without touching live systems.