By Hamza Faisal

A new startup from one of the biggest names in autonomous driving wants to attack the AI industry's most painful bill: compute costs. Tel Aviv-based doubleAI, co-founded by Mobileye co-founder Amnon Shashua, has emerged from two years of stealth to launch what it calls an entirely new category of artificial intelligence — "code execution models," or CEMs — designed to squeeze far more work out of the GPUs and CPUs companies already rent, the Wall Street Journal reported on Monday, October 19.

The idea sits "between code and silicon," in the company's phrasing. Unlike large language models and coding agents that help humans write or modify code, CEMs live inside running programs and operate without human involvement. The debut model, called Maxwell, continuously analyzes a running program, the hardware it runs on, and the data it processes, then optimizes the underlying machine instructions to fit those conditions — without changing the original source code. Each verified improvement can then serve as the basis for further optimization as the program keeps running.

The central technical challenge, Shashua told the Journal, was verification: how to prove the optimized code is still correct. He said the breakthrough was discovering a way to formally and mathematically perform that verification, turning automated code optimization from a risky trick into something customers can trust with production software.

A $220 million bet from Nvidia, Dell and top VCs

Investors are clearly convinced. DoubleAI has raised $220 million to date, according to the Journal, including a Series A round co-led by Lightspeed Venture Partners and Bessemer Venture Partners, with participation from Dell Technologies Capital, Nvidia, Greenfield, Pitango and BRM. Israeli outlet Calcalistech reports the round values the roughly 40-person company at more than $1 billion.

The company is already spending like it means to scale: about half of the raise went toward buying 1,000 Nvidia B300 chips for its own infrastructure. Shashua, who still leads Mobileye — the company said in July he would step down once a successor is named — is now running doubleAI alongside Mobileye colleague Shai Shalev-Shwartz.

Perhaps the most telling detail of the business model: doubleAI plans to charge customers based on the computing cost savings it delivers. It is selling efficiency, and pricing itself against the efficiency it creates.

Why the timing looks right

Compute prices have surged as demand outpaces supply. The hourly cost of renting computing power on Nvidia H100 chips under a one-year contract has climbed 60% in the past year, according to SemiAnalysis. "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."

DoubleAI also positions itself against the AGI hype cycle. The company calls its approach "Artificial Expert Intelligence" — deep domain-expert systems rather than general-purpose intelligence — and claims its models have already demonstrated tricks like solving a 38-year-old graph-theory problem for under $2,000 of compute. The pitch is that the next AI windfall won't come from bigger models, but from software that makes today's hardware run smarter.

The burden of proof is on the benchmarks: doubleAI must now show that mathematically verified self-optimization delivers savings big enough to justify a price tied to the savings. If it can, the company's timing — amid record compute prices — could hardly be better.