Mercor, which supplies expert-written data used to train AI models, announced on 9 July 2026 that it was acquiring Deeptune, a startup that builds simulated copies of business software in which AI agents can practise tasks. SiliconANGLE reported the deal the same day.
Financial terms were not disclosed. Mercor's chief executive, Brendan Foody, announced the deal in a post on the company's blog.
Deeptune's founder and chief executive, Tim Lupo, and the whole Deeptune team are joining Mercor as it expands its New York team, according to Foody's post. Reports differed on where Deeptune was based: SiliconANGLE described it as a New York company, while StartupFortune called it a San Francisco startup.
What Deeptune built
Deeptune builds what it calls training gyms: reinforcement learning environments that reproduce the software knowledge workers use every day. Lupo has compared them to a flight simulator, in which an agent can try out work in replicas of tools such as Excel, Salesforce or Slack without touching a company's real systems.
Over the past two years the company has recreated hundreds of enterprise applications, from spreadsheets to Salesforce, Foody wrote. Its environments were already used by frontier AI labs, and Mercor itself was a Deeptune customer before the deal.
Foody described Deeptune as a small, focused team of engineers and operators with close ties to frontier research. Its headcount was not disclosed.
“It was in a lot of ways the main motivation, actually.”
Why Mercor is buying
Foody's post set out a training environment as three parts: the software the agent works in, the tasks it is asked to do, and the verifiers that check whether it did them well. Mercor already provides the second and third. Its network of more than five million domain experts writes tasks and grading criteria, and it runs the APEX benchmarks that score model performance. Deeptune supplies the first.
Foody argued that the limit on progress in AI training has moved to the environments themselves, and said the combination would make it possible to build realistic environments across many more industries, roles and workflows. For AI labs, he said, environments should ship faster; for enterprises, Mercor wants to be a single partner that builds the environment, staffs the experts and judges whether an agent is ready for real work.
Analysts at Futurum Group noted that few competitors offer both expert-sourced data and simulation, and that Scale AI remains much larger in the AI data services market.
The founder's check
Deeptune raised a $43 million Series A in March 2026 led by Andreessen Horowitz. Foody wrote a personal angel check into that round, nearly four months before the acquisition, SiliconANGLE reported. He told Fortune the investment was made with a takeover already in mind. Foody wrote on Mercor's blog that he first met Lupo about two years earlier and became an early investor.
Other investors in Deeptune were not named in the reports, and the company's founding year was not disclosed.
Mercor's rapid growth
The deal came during a period of fast growth for Mercor. Bloomberg reported that the company was in early talks to raise money at a valuation of about $20 billion, double the $10 billion valuation it reached in October when it closed a $350 million Series C, SiliconANGLE noted. Foody said on X that Mercor's annualised revenue passed $2 billion in June, up 100% in four months.
Mercor has also dealt with a security incident. In March, attackers used malware planted in the open-source LiteLLM library to reach its systems, and the Lapsus$ group later claimed to hold four terabytes of Mercor data. Contract workers have since sued the company. Foody told Fortune the breach had not cost Mercor business and that every frontier lab had expanded its relationship with the company since then.
