Nvidia-Backed Startup DoubleAI Launches Model to Pare Cost of Compute

DoubleAI, a startup based in Tel Aviv, is set to debut what it calls a new category of artificial intelligence that can help companies lower their computing costs.

Co-founded by Chief Executive Amnon Shashua, who also helped launch autonomous driving company Mobileye, doubleAI has the backing of investors that include Lightspeed Venture Partners, Bessemer Venture Partners, Nvidia and Dell, and has raised $220 million total. Shashua is also CEO of Mobileye, but in July the company said he would step down once it names a successor.

DoubleAI describes its new category of AI as “code execution models,” or CEMs, which it says put the AI “between code and silicon.” Unlike large language models that can help humans write code, CEMs live inside running programs and operate without human involvement.

In other words, these AI models help run software and can automatically figure out how to make the code work better. The big challenge in making CEMs work, Shashua said, was determining how to verify the code. DoubleAI’s breakthrough was discovering a way of formally and mathematically performing that verification, he added.

The two-year-old startup’s first CEM, called Maxwell, aims to tackle the problem of soaring compute costs for running graphics processing units, or GPUs, and central processing units, or CPUs, by improving their performance.

“More hardware is being built, but the need for compute is rising faster than the ability to build GPUs and CPUs,” Shashua said. “If you can utilize the hardware better, then you can achieve something incredible in terms of savings.”

Compute prices have surged amid short supply and high demand. The hourly cost of renting computing power from Nvidia H100 chips on a one-year contract has risen 60% in the past year, according to SemiAnalysis, a data and consulting firm.

DoubleAI, which has about 40 employees, plans to charge customers based on how much they save on compute. For instance, if it helps a customer reduce its compute bill from $10 million to $6 million, it keeps some percentage of the $4 million in savings, Shashua said without specifying how much.

The startup said it cut one early customer’s compute costs by 67%.

LTX, an AI company that builds open-weight multimodal models, found that doubleAI’s Maxwell sped up its own model inference twofold.

“The speedup directly translates to cost,” said Zeev Farbman, LTX’s co-founder and CEO. “We measure GPU times, and GPU times have a certain market price at every point in time. Halving the GPU time immediately translates to your serving costs, your margins.”

After doubleAI’s Maxwell becomes more available later this month, Shashua expects that companies with massive compute bills for training and inference—such as AI labs like OpenAI and Anthropic—will see the most benefit.

“There are certain industries or certain companies that don’t care because their compute bill could be $2 million. If they can save half a million dollars, they have bigger fish to fry,” he said. But for the major AI labs, even a 10% compute savings could represent hundreds of millions of dollars, he added.

To solve its own computing needs, doubleAI used roughly half of its funding to purchase 1,000 Nvidia B300 chips. Like the rest of the tech industry, which is facing an urgent compute shortfall, Shashua said the startup didn’t want to scramble for GPUs.

“You cannot rely just on renting compute,” he said. “Prices are going up.”