Nvidia has agreed to invest around five billion dollars in Safe Superintelligence, a private artificial intelligence (AI) company led by Ilya Sutskever, Financial Times reports (open copy). Sutskever previously co-founded OpenAI and helped create the technology behind ChatGPT. The deal forms part of a long-term partnership announced this week. Under the arrangement, Safe Superintelligence will use Nvidia’s newest Vera Rubin hardware. Vera Rubin is Nvidia’s latest generation of specialised computing systems built for artificial intelligence work.
The investment is described as substantial, though exact terms were not confirmed by either company. Such large sums are often paid in stages once certain targets are met. The money will allow Safe Superintelligence to expand its available computing power ten times over the next twelve months. Sutskever said the company has research worth scaling up and that access to a large Nvidia computer will make this possible. Nvidia chief executive Jensen Huang praised Sutskever for earlier fundamental advances in modern artificial intelligence and expressed interest in what new results might emerge from the partnership.
A different research path
Safe Superintelligence still employs only a few dozen people and has not released any product or published research. It was valued at thirty-two billion dollars after raising two billion dollars last year from various investors. Nvidia already held a stake. The company is pursuing a new method of artificial intelligence research that differs from the large language models developed by OpenAI, Google and Anthropic. Sutskever has said the traditional approach of simply adding more data and computing power is producing smaller gains, and that the field is returning to an age of fresh research.
The investment continues a pattern in which chipmakers put money into artificial intelligence developers. Nvidia has made similar large commitments to other customers in the past. Sutskever left OpenAI after a boardroom dispute and founded Safe Superintelligence in 2024 with the single goal of creating safe superintelligence.