Nvidia, Dell join $200 million round for doubleAI

DoubleAI's funding arrives as investors place large bets on startups that can make AI workloads cheaper to run. The company's Maxwell model takes an unusual approach: rather than editing source code, it optimizes the binary that processors actually execute, allowing changes even while applications are running in production. That method has already produced measurable results, including a 38x speedup on a data-science benchmark and a doubling of video-generation speed for a world model called LTX-2.5.

The involvement of Nvidia and Dell is notable because both companies sell the hardware that AI software runs on. DoubleAI plans to spend half the round on 1,000 of Nvidia's B300 Blackwell graphics cards, which the company says will help it compete with OpenAI and Anthropic in automating research tasks. Those rivals have recently shown similar capabilities, with OpenAI releasing 722 AI-generated math papers and an Anthropic model producing a computer-verifiable version of a famous proof.

Beyond commercial optimization, DoubleAI has used its technology for scientific discovery, including an internal agent that made a new finding about mathematical graphs. The chess record, a position with 63 checking moves versus the previous 60, offers a public demonstration of the model's ability to search complex spaces. How quickly enterprises adopt binary-level optimization over traditional source-code changes remains to be seen, but the fresh capital gives DoubleAI room to scale its sales effort.