Tel Aviv-based doubleAI releases a code execution model for improving GPU and CPU performance; doubleAI raised $220M, using ~50% to buy 1,000 Nvidia B300 chips

Your cloud compute bills could fall by up to 67% if your AI workloads can run this new model.

Tel Aviv-based startup doubleAI has released Maxwell, a code execution model (CEM) designed to improve GPU and CPU performance and reduce compute costs for AI operations. Unlike large language models that assist in writing code, Maxwell operates within running programs to optimize their execution. Co-founder Amnon Shashua highlighted that a key challenge was code verification, which doubleAI claims to have solved through formal mathematical methods. The model aims to address the rising expense of compute power by enhancing hardware utilization. Early testing with LTX showed Maxwell doubling inference speed. DoubleAI plans to charge clients a percentage of the compute savings achieved, with potential savings of 10% for major AI labs amounting to hundreds of millions of dollars. The company has raised $220 million, with approximately half allocated to purchasing 1,000 Nvidia B300 chips for its own compute needs, reflecting a strategy beyond just renting hardware.

The introduction of doubleAI's Maxwell CEM signals a significant market shift towards optimizing existing compute resources rather than solely relying on increased hardware production. This approach directly addresses the escalating costs of AI computation, particularly for large-scale training and inference, potentially creating a competitive advantage for companies that can leverage such optimization tools. The startup's value proposition, based on sharing compute savings with clients, suggests a novel revenue model that aligns its success directly with customer efficiency gains.

This development could compel cloud providers and hardware manufacturers to adapt their pricing and service models in response to tools that demonstrably reduce the need for raw compute power. For AI companies, the ability to cut operational expenses by nearly two-thirds could accelerate development cycles and the deployment of more complex models. Investors may see increased interest in companies focused on compute efficiency and optimization software as the demand for AI compute continues to outpace supply.