Artificial intelligence startup Deeptune Inc. has raised $43 million in Series A funding to tackle the AI industry’s data exhaustion crisis. The round was led by Andreessen Horowitz, with participation from 776, Abstract Ventures, Inspired Capital, and angel investors including OpenAI’s Noam Brown and Mercor.io CEO Brendan Foody.
The New York-based company is building high-fidelity virtual "training gyms" for AI agents-simulated workspaces that replicate the digital environments of professionals like DevOps engineers, accountants, and customer support reps. CEO Tim Lupo compares them to flight simulators, enabling AI to learn complex, multi-step tasks through hands-on reinforcement learning.
Instead of relying on scraped web data, Deeptune’s platform runs automated "rollouts" where AI agents practice using tools like Salesforce and Slack, receiving rewards for correct actions. This generates high-quality, scalable training data critical as public internet data nears depletion.
Andreessen Horowitz partner Marco Mascorro praised Deeptune’s impact, noting significant gains in computer-use benchmarks. Recent models now surpass human performance on tasks like OSWorld, with Opus 4.6 scoring 72.7% and GPT-5.4 reaching 75%.
The global reinforcement learning market is projected to grow from $11.6 billion in 2025 to over $90 billion by 2034. Deeptune claims it has already built hundreds of simulations for leading AI labs, proving most human-performed professional tasks can be replicated and mastered by AI.
Mascorro called Deeptune a pioneer in shifting AI training from datasets to engineered environments: “The next decade of AI progress will be driven by better environments.”
Lupo plans to use the funds to expand the team from 20 engineers and researchers, emphasizing New York’s strategic advantage in attracting top-tier AI talent.