Nvidia has signed a $6 billion deal, equivalent to about R$ 30.9 billion, with the startup Poolside. This deal involves licensing Poolside's technology and hiring most of its engineers, aiming to accelerate the development of open-weight artificial intelligence models capable of competing with systems developed in China.
According to The Wall Street Journal, this move positions Nvidia in direct competition with giants like OpenAI and Anthropic. The company, which traditionally supplies the chips used by major AI labs, is now expanding its activities into developing its own models.
The agreement also includes a $1 billion (R$ 5.16 billion) financial injection into Poolside, whose previous valuation was $12 billion (R$ 62 billion). More than one hundred employees from the startup, mostly engineers, will be integrated into Nvidia. This group will work on Nemotron, an open-weight model project that the company presented in 2023.
Poolside was founded in 2023 by Eiso Kant and Jason Warner, former CTO of GitHub, and is led by Margarida Garcia. These three individuals will remain outside of Nvidia, continuing involvement in undisclosed research.
The main distinction of open-weight models lies in the ease of download and modification, which allows systems to be adapted to various needs and generally operate with reduced costs.
Nvidia intensified its focus on this segment after the notable progress of Chinese models. In March, the company established the Nemotron Coalition, bringing together companies such as Mistral, Thinking Machines Lab, and Perplexity, with the aim of sharing knowledge, data, and computational resources.
A milestone that intensified pressure on American corporations was the launch of the first low-cost open-weight model by DeepSeek, which occurred in January 2025.
Among the cited Chinese models, Jensen Huang, CEO of Nvidia, expressed on an X post that American leadership will depend on creating an open ecosystem capable of reaching multiple sectors.
Poolside's financial situation also facilitated the merger of the two companies. At the end of 2025, the startup sought to raise $2 billion (R$ 10.3 billion) over a six-week period to finance a cluster composed of 40,000 GB300 servers, but the fundraising was not completed.
Poolside's founders informed shareholders that 'we did not close in time and lost the cluster.' This incident highlighted another vulnerability: without new resources, the company risked exhausting its computational capacity as early as the following year. Nvidia possessed exactly the infrastructure and capital needed to mitigate this restriction.
Shortly before the deal, Poolside had introduced Laguna S, a model that gained prominence among open-weight systems in the West.
This operation places Nvidia in a unique position. Although OpenAI and Anthropic are major customers of its chips, they also develop models that can directly compete with Nemotron.
Advanced Development and Strategic Shift
Currently, the company is working on the most sophisticated version of the project, which has the potential to exceed one trillion parameters. It is worth noting that some Chinese models already have larger sizes.
With the acquisition of Poolside, Nvidia transcends the role of mere backstage support in the AI race, beginning to actively participate in the development of models that rely on its own processors.