DoubleAI’s “Maxwell” aims to run code directly with AI, while Nvidia backs the idea that software will erode the need for bespoke chips

Nvidia is backing a young startup whose pitch runs counter to the instinct of many chip makers: make advanced computing less dependent on ever-more specialized hardware. According to a Yahoo Finance report, the company is investing in doubleAI, a Tel Aviv-based firm developing a system it frames as a “code execution model,” an approach that focuses on having AI models carry out tasks by writing and running code-like actions rather than relying only on text responses.

The report ties Nvidia’s investment to a larger shift in how AI applications may be built. In the framing described by Yahoo Finance, doubleAI’s goal is to lower the effort and cost required to get working outcomes from AI, which could, in turn, reduce the premium that customers pay for highly optimized, purpose-built compute stacks. The implication is not that chips stop mattering, but that the value chain may move from hardware differentiation toward software and workflow abstraction.

Central to doubleAI’s pitch, as characterized in the report title and description, is “Maxwell,” which the startup positions as its first “code execution model.” A code execution model, in plain terms, is meant to do more than generate instructions. It aims to translate an intent into executable steps, often by producing code that an application can run, validate, or otherwise carry out as part of a larger automation loop. That matters because enterprises typically want results they can verify and integrate, not just chat outputs.

Yahoo Finance also reports that Nvidia is participating in a $220 million funding round that would bring capital to doubleAI. The company’s involvement indicates that at least some of Nvidia’s strategy may include funding the platforms that consume its compute, even if those platforms eventually standardize or commoditize parts of the stack. For Nvidia, investing in tooling that drives more AI workloads can still be a business win, even if the tooling reduces the need for specialized configurations.

doubleAI’s model approach arrives as the market debates how AI systems should be measured and deployed. Systems that can execute code may be easier to evaluate through outcomes like correctness, completion rate, and reliability. That can be attractive to developers trying to operationalize AI, including teams building agentic workflows where an AI system repeatedly plans, attempts an action, and corrects itself based on observed results.

What Nvidia did not disclose, at least in the material available for this story, is any detail about the specific economics of the investment, such as the amount Nvidia personally put in, the structure of its stake, valuation terms, board rights, or whether the company will provide technical support. The publicly available information here also does not spell out whether doubleAI is targeting any particular customer segment, what infrastructure the company expects to run on, or how Maxwell compares with existing “agent” or tool-use systems in benchmark terms.

Looking ahead, investors and developers will likely watch two questions. First, whether Maxwell-style “code execution” delivers measurable improvements in task reliability and verification, not just demonstrations. Second, how Nvidia’s backing fits into its broader ecosystem approach, where software abstractions may become more important over time while compute demand still expands through increased usage and new classes of automation.