Nvidia, Dell anchor $200M round for AI optimization startup doubleAI

Tel Aviv-based doubleAI, officially AA-I Technologies, has raised $200 million in a Series A round co-led by Lightspeed and Bessemer, with strategic investments from Nvidia and Dell. The round values the company at over $1 billion, according to reporting by CTech.

doubleAI’s flagship product, an AI model called Maxwell, specializes in automatically optimizing software binaries to reduce hardware usage. Unlike traditional compilers or manual tuning, Maxwell analyzes how a program interacts with data and underlying chips—and applies optimizations directly to the binary, allowing changes to be made while applications run in production. SiliconANGLE reported that the model has already demonstrated tangible results: it sped up 55% of the algorithms in Nvidia’s open-source graph-analytics library, cuGraph, by a factor of two or more. The model has also shown significant performance improvements for widely used data-science tools, including a recent project where it enhanced video-generation speeds for a world model.

The company plans to sell Maxwell to enterprises looking to lower the hardware costs of their AI workloads. This positions doubleAI in a growing niche where startups compete not only on model capabilities but on efficiency gains that can translate into direct cost savings. The involvement of Nvidia and Dell is particularly notable. SiliconANGLE reported that doubleAI will use half of its $200 million funding to purchase 1,000 of Nvidia’s B300 Blackwell GPUs, a move that aligns with Nvidia’s broader strategy of embedding its hardware deeper into the AI stack. For Dell, the investment extends its recent pivot from selling GPUs to offering end-to-end enterprise AI solutions, a shift highlighted during its AI Leadership Symposium last week.

Beyond optimization, doubleAI has demonstrated Maxwell’s versatility in research settings. Earlier this month, the model broke a nearly 50-year-old chess record by finding a position where 63 moves give check, surpassing the previous record of 60 moves. The company has also used its AI agents to make advances in mathematical research, including discoveries that can be easily verified through code-based proofs.

These achievements reflect a broader trend where AI models are increasingly automating complex research tasks, a space currently dominated by players like OpenAI and Anthropic. The Wall Street Journal noted that doubleAI’s focus on efficiency could help it carve out a distinct competitive edge, particularly as enterprises grapple with the rising costs of running large-scale AI workloads.

The funding round arrives amid heightened investor interest in AI infrastructure startups that promise to reduce the operational overhead of deploying and scaling models. When we covered Nvidia’s open-sourcing of its AI agent safety platform last month, the emphasis was on managing risks posed by autonomous agents. doubleAI’s approach, by contrast, targets performance. Its ability to optimize binaries without requiring source-code changes could appeal to enterprises that rely on proprietary or legacy software, where manual tuning is either impractical or impossible.

The $1 billion valuation and the participation of two semiconductor heavyweights suggest confidence in doubleAI’s technical approach, but questions remain about its scalability and market adoption. While Maxwell has shown impressive benchmarks, its enterprise sales strategy is still unproven. The company’s decision to allocate half its funding to hardware purchases—rather than go-to-market efforts—indicates a bet on both its own technology and Nvidia’s roadmap. Whether enterprises will prioritize runtime optimization over model scale or feature richness is an open question, and one that will likely shape the next phase of the AI infrastructure race.

Sources: siliconangle.com