“We’re dramatically increasing our hardware support,” a source close to the deal said, highlighting Nvidia’s growing commitment to Safe Superintelligence. The AI startup, founded by Ilya Sutskever, former chief scientist at OpenAI, is set to scale up its computing power substantially thanks to a new partnership with Nvidia. This move marks a significant shift away from the startup’s previous dependence on Google’s tensor processing units (TPUs).
Safe Superintelligence, known for its secretive approach to developing what it calls “safe superintelligence,” has raised nearly $2 billion in funding, reaching a valuation near $30 billion despite lacking any public releases or clear product roadmaps. Nvidia’s investment, described as “substantial” but undisclosed in amount, will provide the AI lab with access to powerful GPUs, which are essential for the massive training workloads that come with advanced AI research.
Previously, Safe Superintelligence had relied mainly on Google’s processors, making this partnership with Nvidia a strategic attempt to diversify its hardware suppliers. Nvidia’s chips, widely used for deep learning, could boost the startup’s computational capacity by an order of magnitude, accelerating its research efforts. This collaboration follows a similar Nvidia deal earlier this year with Mira Murati’s Thinking Machines Lab, underscoring Nvidia’s push to capture elite AI research clients as competition intensifies.
Formed in 2024, Safe Superintelligence seeks to build AI systems capable of human-level tasks without risk to humanity, focusing on long-term, high-stakes development over incremental product releases. The firm’s backers include major venture capital names like Andreessen Horowitz and Sequoia Capital. While Nvidia has yet to announce specific timelines for hardware delivery, the alliance signals a deepening rivalry with Google in the AI chip market and research landscape.
