Tel Aviv startup doubleAI is debuting Maxwell, the first of what it calls “code execution models” — AI that lives inside running software and makes it run more efficiently on GPUs and CPUs, rather than helping humans write code. Why it matters: with compute prices surging, squeezing more work out of existing chips could save the biggest AI labs hundreds of millions of dollars.
October 19, 2026
What happened?
DoubleAI, a two-year-old startup based in Tel Aviv, is set to debut its new category of artificial intelligence, the Wall Street Journal reported on October 19. The company was co-founded by chief executive Amnon Shashua, who also helped launch autonomous driving company Mobileye. Its investors include Lightspeed Venture Partners, Bessemer Venture Partners, Nvidia and Dell, and it has raised $220 million in total. Shashua is also CEO of Mobileye, though the company said in July he would step down from that role once a successor is named.
What is a code execution model?
DoubleAI describes code execution models, or CEMs, as putting AI “between code and silicon.” Unlike large language models that help humans write code, CEMs live inside running programs and operate without human involvement — automatically figuring out how to make the code work better.
The hard part, Shashua said, was verification: proving the optimized code is correct. DoubleAI says its breakthrough was finding a way to perform that verification formally and mathematically.
What does Maxwell claim to do?
Maxwell, the company’s first CEM, targets the soaring cost of running GPUs and CPUs by improving their performance. The startup said it cut one early customer’s compute costs by 67%. LTX, an AI company that builds open-weight multimodal models, said Maxwell sped up its own model inference twofold — and, as LTX co-founder and CEO Zeev Farbman put it, halving GPU time translates directly into lower serving costs and better margins.
The business model matches the pitch: doubleAI, which has about 40 employees, plans to charge customers based on how much they save on compute, keeping a percentage of the savings. It also spent roughly half its funding buying 1,000 Nvidia B300 chips for its own computing needs, rather than relying on rentals as prices rise. For context, SemiAnalysis data cited by the Journal puts the rise in one-year H100 rental costs at 60% in the past year.
Why it matters
Most AI efficiency stories are about building more hardware. DoubleAI is betting the cheaper win is using the hardware that already exists more intelligently — and pricing itself as a share of the savings rather than another line-item cost. Shashua expects the biggest beneficiaries, once Maxwell becomes more available later this month, to be companies with massive training and inference bills such as OpenAI and Anthropic, where even a 10% saving could mean hundreds of millions of dollars. The claims so far rest on the company’s own early-customer results, not independent benchmarks, so they need testing at scale. But the idea — AI that optimizes the machines AI runs on — is a genuinely new category worth watching.
Frequently asked questions
What is doubleAI?
DoubleAI is a two-year-old Tel Aviv startup co-founded by Amnon Shashua. It has raised $220 million from investors including Lightspeed, Bessemer, Nvidia and Dell, according to the Wall Street Journal.
How is a code execution model different from a large language model?
An LLM helps humans write code. A code execution model lives inside running programs, operates without human involvement, and automatically makes that code run more efficiently on GPUs and CPUs.
How will doubleAI make money?
The company plans to charge based on how much customers save on compute, keeping a percentage of those savings.
Sources: The Wall Street Journal