英伟达据报以6亿美元获得Poolside模型构建系统的非独家授权,并向其109名员工提供职位,另以120亿美元投前估值向Poolside投资10亿美元,创始人留任。该授权针对Poolside内部训练、评估、强化学习及合成数据生成的Model Factory平台,而非Laguna成品模型。英伟达借此获得更快的研究循环,Poolside称原本需数周排期的实验现可在一小时内运行。
Nvidia reportedly agreed to pay $6B for a non-exclusive license to Poolside’s model-building system.
Nvidia is also offering jobs to 109 employees and investing $1B in Poolside at a $12B pre-money valuation, while Poolside’s founders stay.
The licensed Model Factory is Poolside’s internal platform for training, evaluation, reinforcement learning, architecture testing, synthetic data generation and data mixing across GPU clusters. Nvidia's $6B license targets this research process of Poolside, rather than a finished Laguna model.
An by licensing Model Factory, Nvidia is buying a faster research loop that may matter more than any one Laguna release.
Poolside says experiments that once took weeks to schedule can run in under an hour, giving Nvidia access to a much faster model-research loop.
The deal structure itself may become a repeatable way for Nvidia to absorb scarce AI capability without buying entire startups. It has reportedly used similar licensing-and-hiring arrangements with Groq and Enfabrica
来源:Rohan Paul · x.com