DoubleAI, a Tel Aviv-based startup backed by Nvidia $NVDA, has launched an AI model that it says lowered the compute spending of one early customer by 67%, according to The Wall Street Journal.
The model, called Maxwell, is the company's first "code execution model," or CEM. It is slated for broader release before the month is out.
Whereas large language models assist people in writing software, CEMs sit within running programs, work autonomously, and figure out ways to improve how the code performs. Co-founder and CEO Amnon Shashua said verifying the code was the main obstacle. DoubleAI found a way to perform that verification formally and mathematically, he said.
Maxwell is meant to ease the growing expense of operating GPUs and CPUs, doing so by getting more out of that hardware.
"More hardware is being built, but the need for compute is rising faster than the ability to build GPUs and CPUs," Shashua told The Wall Street Journal. "If you can utilize the hardware better, then you can achieve something incredible in terms of savings."
The hourly price to rent Nvidia H100 chip capacity under a one-year contract is up 60% from a year earlier, according to SemiAnalysis, a consultancy that focuses on data.
LTX, a maker of open-weight multimodal AI models, found that Maxwell made inference for its own model twice as fast. “We measure GPU times, and GPU times have a certain market price at every point in time,” Zeev Farbman, LTX’s co-founder and CEO, told The Wall Street Journal.
DoubleAI intends to bill customers according to the size of the savings they see. Shashua said that if the company brought a customer's spending on compute down from $10 million to $6 million, it would collect a share of the $4 million saved. He did not say what percentage.
Shashua expects companies with the largest training and inference bills, such as OpenAI and Anthropic, to benefit most once Maxwell is more widely available. He said a 10% savings could be worth hundreds of millions to the largest AI labs.
DoubleAI has raised $220 million in total. Its investors include Lightspeed Venture Partners, Bessemer Venture Partners, Nvidia and Dell $DELL. The two-year-old company has about 40 employees. Shashua is also CEO of Mobileye.
The startup put about half the money it raised toward buying 1,000 Nvidia B300 chips for its own computing needs. "You cannot rely just on renting compute," Shashua said, according to The Wall Street Journal. "Prices are going up."