Mercor buys Deeptune, a startup its CEO angel-funded

Mercor, the AI talent and data platform, said on July 9 that it will acquire Deeptune, a New York startup that builds the reinforcement-learning environments where AI agents rehearse real work. The full team, led by founder-CEO Tim Lupo, joins Mercor and expands its New York presence. Terms were not disclosed.

The deal is small on paper and loud in what it signals. Mercor argues that progress in agentic AI is now bottlenecked by high-quality training environments, not raw model capability, which makes the firms that build those environments the picks-and-shovels targets of the moment, a bet now playing out across the business-tech landscape.

What’s actually new

Deeptune has recreated hundreds of enterprise applications, from spreadsheets to Salesforce, so agents can practice tasks in a simulated copy before touching production systems. Mercor brings the expert network and the task and evaluation design; Deeptune brings the simulated apps. Together that is a full loop for training and grading agents on enterprise work.

The timing is the unusual part. Deeptune raised a $43 million Series A led by Andreessen Horowitz in March, roughly three months before this deal, with participation from 776, Abstract Ventures, Inspired Capital, and angels including OpenAI’s Noam Brown. One of those angels was Mercor CEO Brendan Foody himself. Foody, 23, told Fortune that the investment was “in a lot of ways the main motivation” for the acquisition, an unusually direct, on-record admission that he backed the company with a purchase in mind. The rationale is laid out in Mercor’s announcement.

Mercor is negotiating from strength. The Information reported in June that the company hit $2 billion in gross annualized revenue, up from about $1 billion a year earlier and roughly $1 million 24 months ago, a trajectory that fits the new pattern of AI startups reaching nine-figure revenue in record time. Read that top-line figure carefully: it is gross run-rate billings before Mercor pays its experts and contractors, who reportedly keep 60 to 70 percent, which puts net revenue closer to $600 to $800 million by outside estimates. TechCrunch reported the same week that Mercor is in talks to raise at roughly a $20 billion valuation, double its last mark of $10 billion, though those talks are not closed.

Mercor is acquiring @deeptuneai.

Training AI to do real professional work requires covering the full distribution of apps, worlds, and tasks across the economy. We have built the largest expert network and have been a leader in creating worlds and tasks. Deeptune built the apps,…

— Mercor (@mercor_ai) July 9, 2026

What this means for teams buying agentic AI

If you are evaluating agents for real workflows, this deal is a tell about where reliability comes from. Vendors are consolidating the training-and-evaluation stack so agents can be drilled and scored on your kind of work before they run it. That is the same lock-in dynamic that surfaces once agents leave the pilot phase: whoever owns the environments and the evals owns the switching costs. Mercor is not alone in moving early; the agent race has broadly turned to M&A.

Foody also claims every “Magnificent Seven” company except Tesla is now a Mercor customer, a company figure Fortune reported but did not independently verify. Treat it as a vendor claim, not a checked fact, when it lands in a sales deck.

What to watch

Three things. Whether the $20 billion round closes at that number or lands lower, which will test how much the market actually believes the environments thesis. Whether Foody’s invest-then-acquire playbook becomes a repeatable pattern, given how cheaply environment builders can be seeded and then absorbed. And whether Mercor’s reported March data breach, tied to a LiteLLM vulnerability and disputed by the company on severity, resurfaces as it takes on more enterprise simulation work. The signal is clear even if the price is not: the agent economy’s next moat is the practice field, not the model.