O que aconteceu
Global Times reports that The Wall Street Journal said Nvidia will pay $6 billion to license artificial-intelligence model-building technology from Poolside and hire more than 100 of the startup’s engineers. The reported goal is to build an open-weight model that can compete with Chinese systems including DeepSeek and Moonshot AI’s Kimi K3. Global Times also says Nvidia will separately invest $1 billion in Poolside at a $12 billion pre-money valuation, with Poolside’s co-founders remaining at the company.
Global Times reports that The Wall Street Journal said Nvidia will pay $6 billion to license Poolside’s model-building technology and hire more than 100 Poolside engineers. According to the report, Nvidia intends to use that technology and team to build an open-weight AI model capable of competing with Chinese models such as DeepSeek and Moonshot AI’s Kimi K3. The article says Nvidia will also invest $1 billion in Poolside separately, at a $12 billion pre-money valuation, while Poolside’s co-founders remain with the company. These deal terms are reported by Global Times through its account of the Wall Street Journal report and are not independently confirmed here.
Global Times says Nvidia CEO Jensen Huang framed the effort in national terms in a letter titled “Open Weights and American AI Leadership.” The article reports that Huang argued the AI race should be judged not by a single frontier model but by whether the United States develops an open ecosystem that spreads across sectors. The source does not provide the full letter, a release date for Nvidia’s proposed model, technical specifications or details of the licensing agreement. It therefore establishes a reported strategic intention, not a demonstrated product launch or tested model.
The article’s central interpretation comes from two analysts quoted by Global Times. Chen Jing of the Technology and Strategy Research Institute describes the move as a strategic adjustment under competitive pressure, while Xiang Ligang of the Zhongguancun Modern Information Consumer Application Industry Technology Alliance says Chinese companies have created structural anxiety for Nvidia. Global Times also cites reported market indicators: Chinese open-weight models accounted for about 41 percent of Hugging Face downloads over the prior year, according to the article; Bloomberg reported that Airbnb, DoorDash and Coinbase adopted Chinese models hosted on local servers; and Bernama reported deployments involving Qwen in Singapore and DeepSeek in Malaysia. Those figures and deployments are attributed to the cited outlets rather than independently verified by Global Times in the supplied text.
Leia a fonte primária: globaltimes.cn ↗
Por que isso importa
The report presents the transaction as a significant strategic move by a chipmaker into the model layer. Analysts quoted by Global Times say Nvidia has an economic incentive to spread capable models widely because broader adoption can increase demand for training and inference hardware. They also argue that Chinese open-weight models have gained attention and usage despite US restrictions on advanced chip exports, putting pressure on the closed-model strategy favored by major US AI companies.
Global Times’s analysts argue that Nvidia’s position gives it a distinct commercial reason to support open-weight AI. If capable models are easier to obtain and adapt, more organizations may experiment with them, which could expand demand for the GPUs and related infrastructure Nvidia sells. That is an economic explanation for the reported strategy, not evidence that the deal will increase Nvidia’s sales or that the proposed model will outperform alternatives. The article provides no financial forecast, customer commitment or independent assessment of the business case.
The report also places the deal in a contest over how AI capabilities spread. Global Times says Xiang described Chinese open-weight models as gaining support and users quickly because their weights are available, allowing developers to reuse and improve them. It cites Alibaba’s disclosure that its Qwen models accumulated more than 3 billion downloads in six months, as well as reported use of Qwen in Singapore’s government-backed AI Singapore program and DeepSeek on locally sited servers in Malaysia. The source does not establish how those downloads compare with active usage, how the deployments perform, or whether they involve production-critical decisions.
The strategic distinction matters because Nvidia’s move would not by itself represent a broad conversion of the US model industry. Chen told Global Times that OpenAI, Anthropic and Google still treat model architecture and weights as competitive advantages. The article characterizes Nvidia’s reported effort as a defensive attempt to support a Western open-weight standard-bearer, while noting that the priority could change if competitive pressure eases. It also says differences over data governance, chip controls and standards could continue even if core model architectures remain interoperable. Whether this becomes a durable industry shift depends on actions by model companies beyond Nvidia.
O que assistir a seguir
The reported deal still requires confirmation from Nvidia, Poolside or the Wall Street Journal’s original reporting. Key unknowns include the licensed technology, the scope of Nvidia’s investment, the model’s architecture and license, the timing of any release, and whether the model will be genuinely open-weight in practice. The most consequential follow-up will be whether Nvidia’s strategy produces a widely used model and whether other US model developers respond by releasing more of their own weights.
The first verification point is the transaction itself. Nvidia, Poolside or the Wall Street Journal’s original report could clarify whether the $6 billion figure covers a license, another form of technology transfer, or additional commitments; what the separate $1 billion investment buys; and how many engineers are actually joining Nvidia. The supplied Global Times report offers no primary agreement, regulatory filing or statement from the companies, so those details remain unconfirmed.
The next test is the proposed model. There is no model name, release date, parameter count, training disclosure, benchmark result, access policy or license in the source. It is also not clear whether “open-weight” will mean downloadable weights only or a broader release of code, data information and tools. Those distinctions will determine how independently developers can run, inspect, adapt and redistribute the system.
Finally, observers should watch adoption rather than strategic language alone. Useful evidence would include a public release, independent evaluations against Chinese and US models, meaningful use by organizations outside Nvidia, and signs that the effort changes demand for computing infrastructure. The report also raises a broader question: whether other US companies begin releasing weights, or whether Nvidia’s project remains a company-specific response to Chinese competition. None of those outcomes is established by the current report.
