Nvidia to invest $6 billion to build U.S. open-weight AI model alternative to China’s DeepSeek, Kimi K3, and Qwen

Nvidia is committing roughly $6 billion to a sweeping deal with AI startup Poolside, aiming to build one of the most capable U.S. open-weight AI models yet and challenge China’s growing lead in downloadable models from DeepSeek, Moonshot AI, and Alibaba’s Qwen.

The deal marks a significant shift for Nvidia. The company that became the backbone of the AI boom by selling GPUs is moving deeper into the model layer, backing an American open-weight contender with its money, computing infrastructure, and engineering resources.

The Wall Street Journal reported that Nvidia’s agreement with Poolside includes about $1 billion invested in the startup, access to its technology, and the transfer of more than 100 Poolside engineers to Nvidia’s Nemotron model effort. Poolside’s founders are expected to remain at the startup rather than join Nvidia.

“Nvidia is planning to use a $6 billion deal it struck this week to build one of the world’s most powerful open-weight AI models, one that would compete with Chinese heavyweights like DeepSeek and Kimi K3,” the Wall Street Journal reported, citing people familiar with the matter.

That puts the scale of Nvidia’s ambitions into perspective. This is bigger than another strategic investment in an AI startup. Nvidia appears to be assembling the technology, talent, and infrastructure needed to compete directly at the foundation-model layer.

China’s open-weight AI push is changing the race

The timing matters. Chinese AI developers have emerged as major forces in open-weight AI, giving developers access to model weights they can download, modify, fine-tune, and deploy on their own infrastructure.

DeepSeek helped bring that shift into the mainstream. Alibaba’s Qwen family has built a large global developer following. Moonshot AI pushed the competition further with Kimi K3, a 2.8-trillion-parameter model whose full weights were released last month.

Those releases have created a different competitive threat for U.S. AI companies. OpenAI and Anthropic largely keep their most advanced model weights closed, giving customers access through products and APIs. Open-weight models give developers far more control and can offer lower operating costs for companies willing to run or customize them themselves.

Nvidia now has a clear incentive to ensure the open-weight side of AI does not become dominated by Chinese models.

Nemotron could become its answer.

The Poolside agreement has another strategic layer. Every capable model built around Nvidia’s software and hardware ecosystem can drive more demand for the company’s GPUs, networking equipment, CUDA software, and AI infrastructure. Nvidia does not need to replace OpenAI or Anthropic for the strategy to work. A thriving ecosystem of open models optimized around Nvidia technology could strengthen the company’s position across the AI stack.

The structure of the Poolside deal reflects another shift taking place across Silicon Valley. Rather than buying promising AI startups outright, large technology companies have increasingly used investments, licensing agreements, and talent deals to gain access to valuable technology and engineers without completing conventional acquisitions.

For Nvidia, more than 100 engineers joining Nemotron could be nearly as important as the technology itself. Frontier AI talent has become one of the industry’s scarcest resources.

Why a U.S. Open-Weight AI Model Matters

Nvidia built a multitrillion-dollar business supplying the infrastructure behind the AI race. Its $6 billion Poolside deal suggests it no longer wants to sit primarily beneath the model makers. Nvidia is using the cash generated by its chip dominance to move higher into the AI stack, just as DeepSeek, Kimi K3, and Qwen are making Chinese open-weight models increasingly difficult for the U.S. AI industry to ignore. If Nemotron becomes a serious open-weight alternative, Nvidia could end up supplying both the machines that run the AI and one of the models running on them.