doubleAI Launches "Code Execution Model" Maxwell With Nvidia Backing; First Customer Cut Compute Costs 67%

What happened

doubleAI was co-founded by Mobileye co-founder and CEO Amnon Shashua and is based in Tel Aviv. The company has raised $220 million in total from investors including Lightspeed Venture Partners, Bessemer Venture Partners, Nvidia and Dell.

Unlike large language models that help humans write code, doubleAI describes CEMs as AI that sits "between code and silicon": it lives inside running programs and optimizes software on its own, without human involvement. Shashua said the biggest challenge in making CEMs work was determining how to verify the code, and the company's breakthrough was finding a way to perform that verification formally and mathematically.

To meet its own computing needs, doubleAI spent roughly half of its funding to purchase 1,000 Nvidia B300 chips for its own use. Shashua said the startup did not want to scramble for GPUs amid the shortage.

Key facts

Early customer data: One early customer cut compute costs by 67%; AI company LTX, which builds open-weight multimodal models, found Maxwell doubled its model inference speed. LTX co-founder and CEO Zeev Farbman said the speedup translates directly into cost savings.

Pricing: doubleAI plans to charge based on a share of the compute savings it delivers. For example, if it cuts a customer's $10 million compute bill to $6 million, it keeps a percentage of the $4 million saved; the exact share was not disclosed.

Target customers: Shashua expects companies with massive compute bills, such as AI labs like OpenAI and Anthropic, to benefit most once Maxwell becomes more widely available later this month. He said that for major AI labs, even a 10% compute saving could represent hundreds of millions of dollars.

Market context: The hourly rental cost of Nvidia H100 chips on one-year contracts has risen 60% in the past year, according to data and consulting firm SemiAnalysis, as short supply and high demand push prices up.