doubleAI has opened Maxwell to software developers in early access, pitching it as a way for companies to lower computing bills without rewriting their applications. The Tel Aviv startup also formally disclosed a $200 million Series A led by Lightspeed Venture Partners and Bessemer Venture Partners.
Most coding assistants work before execution: a person requests code, and the model produces or edits the source. Maxwell works later. It observes a running program together with the processor and data, then changes the machine instructions beneath the application while leaving its source code untouched. doubleAI says it can optimize workloads on both CPUs and GPUs.
Changing live machine instructions is useful only if the output stays the same. doubleAI says Maxwell formally proves that each rewrite preserves the program’s behavior. When the conditions supporting an optimization no longer apply, a guard returns control to the original code, allowing changes to take effect without a person reviewing each one.
The performance figures remain company claims. doubleAI says Maxwell has made data pipelines more than 38 times faster and AI inference kernels more than 4.5 times faster on real-world workloads. CTech noted that the announcement lacks enough detail about tests, hardware configurations and baselines to determine how broadly those results can be repeated.
doubleAI plans to charge customers a share of the computing costs Maxwell saves, although it hasn’t disclosed the percentage. That ties its revenue to reductions in hardware use rather than the number of software seats sold. The initial audience is enterprises and AI labs running CPU- or GPU-intensive workloads.
The financing itself isn’t new. Globes reported in November 2025 that the $200 million round had already closed at a valuation of about $1 billion and that AA-I Technologies, now doubleAI, had raised $221 million in total. The October 19 release is the company’s confirmation alongside Maxwell’s launch, not a second $200 million transaction.