doubleAI Raises $200M to Speed Up Running Code
doubleAI raised $200M from Nvidia, Dell and others to build AI that speeds up running software without touching its code, at a reported $1B+ valuation.
A new AI startup called doubleAI has come out of stealth with a $200 million funding round and an unusual idea: instead of helping programmers write code, its AI makes the code that is already running go faster. The Tel Aviv-based company was founded by Amnon Shashua, who previously led Mobileye, and its backers include Nvidia and Dell.
The $200 million round
doubleAI announced on October 19, 2026 that it raised $200 million in a Series A round, co-led by Lightspeed Venture Partners and Bessemer Venture Partners. Other investors include Dell Technologies Capital, Greenfield, Nvidia, Pitango and BRM. Reports value the company at more than $1 billion, and the company says it employs about 40 people. The company plans to use part of the money to buy more computing hardware, with the Wall Street Journal reporting that about half of the round will go toward 1,000 of Nvidia's B300 Blackwell graphics processors.
AI that lives inside running code
doubleAI says it has created a new kind of AI model called a Code Execution Model, which works "between the code and the silicon" - meaning between the software and the hardware it runs on. Its first model, Maxwell, watches a program while it runs, studies how the program uses data and the underlying processor, and rewrites the machine instructions underneath so the program runs faster. The company says every rewrite is mathematically proven correct, and the original source code never changes. The approach works on both central processors (CPUs) and graphics processors (GPUs).
What the tests showed
To show what Maxwell can do, doubleAI says the model rewrote parts of Nvidia's cuGraph, an open-source library for graph analytics, speeding up 55 percent of its algorithms by two times or more. It also created a faster version of pandas, a popular Python data-science library, improving its score on the PDS-H query-speed benchmark by a factor of 38. In another test, Maxwell doubled the video-generation speed of a world model called LTX-2.5. The company says each proven improvement becomes the starting point for the next one, so a program can keep getting faster the longer it runs.
Why it matters
Big AI models cost a fortune to run because they burn through expensive chips. If Maxwell can really make software run faster without developers changing anything, companies could cut those bills sharply - and doubleAI plans to charge customers based on the computing costs it saves them. The idea comes at a time when companies are racing to cut the cost of running AI, so investors are betting this new layer of AI between code and chips is worth a billion dollars. The catch, as with any young company, is that the headline numbers come from doubleAI's own tests and have not been checked independently yet.