One of the most consequential storylines in AI right now is no longer about a single company or model, but about geography. The race to build the best openly available AI models, the ones whose weights are released freely for anyone to download and run, has become a contest between the United States and China, and China is winning much of it. Chinese labs are releasing a string of open models that match or exceed the best open models from America, upending the assumption that the US holds an unquestioned lead in AI. This geopolitical dimension is reshaping the strategy of every player in the AI stack, including the data platforms, and it is an essential context for understanding the current moment.
Open Weights, Briefly
The distinction is worth stating plainly. A closed model, like the ones powering ChatGPT, is kept behind a company’s service; you can use it, but you cannot download it, inspect it, or run it yourself. An open-weights model is released for anyone to download, run on their own hardware, modify, and build on, freely. The open approach drives rapid, decentralized innovation and lets enterprises run and customize models in their own secure environments, which is exactly why open models matter so much to the data platforms helping companies build private AI.
For a while, the strongest open models came largely from American and European efforts. Now Chinese labs are releasing open models that are not just competitive but, by several measures, the best openly available models in the world, and they are doing so repeatedly, establishing a clear presence at the frontier of open AI.
Why China Is Pursuing Open Models So Aggressively
There is a strategic logic to why Chinese labs lean into open releases. Releasing powerful models openly builds global influence and adoption: developers and companies around the world build on models that are freely available, which spreads the releasing country’s technology, standards, and ecosystem. It is a way to gain ground in a field where the leading closed models are American, by competing on a different axis, openness, rather than trying to out-spend the closed-model incumbents directly. Open releases also blunt the advantage of restrictions, since a freely downloadable model cannot easily be cut off.
This is unfolding against a backdrop of broader US-China technology tension, including American restrictions on exporting the most advanced AI chips to China. Those restrictions are meant to slow China’s AI progress by limiting access to the powerful hardware needed to train frontier models. The competitive Chinese open models suggest that the restrictions, whatever their other effects, have not prevented Chinese labs from producing world-class AI, and may even have spurred greater efficiency and a strategic embrace of openness as a counter.
The American Response
The strength of Chinese open models is prompting soul-searching and response in the American AI world. There is growing concern that if the world’s developers build on Chinese open models by default, American influence over the direction of AI will erode. This is helping motivate renewed American commitment to competitive open models. NVIDIA has released its Nemotron open models explicitly positioned as the most capable open models built in America, backed by a major multi-year financial commitment to open-weight AI development, while being candid that the very best open model overall still comes from China. That framing, the best American open model, with the honest acknowledgment of what it is not, captures the state of the race precisely.
What It Means for the Market
For the data platforms, the open-weights race has direct and practical consequences. Enterprises building private AI on their own data, exactly the use case Databricks and Snowflake are chasing, often prefer open models precisely because they can be run in the company’s own secure environment, customized freely, and not subject to a single vendor’s control or pricing. A vibrant, competitive ecosystem of powerful open models is therefore good for the data platforms, because it gives their customers strong, flexible models to deploy on the platform rather than being forced to route everything through a closed external service. The platforms increasingly offer easy access to a menu of open models, letting customers choose.
The geopolitical dimension also introduces new considerations enterprises have never had to weigh before: which country’s models to build on, what the security and governance implications are, and how shifting restrictions might affect their choices. The benefit of the open-weights race to the world is a flood of increasingly capable freely available models, driving down costs and broadening access to frontier AI, democratization playing out at a global scale. The risks are the harder-to-control nature of openly released powerful models and the entanglement of AI with national rivalry and security concerns. For the data platforms, the supply of AI models, open and closed, American and Chinese, has become a rich and contested landscape, and their value increasingly lies in being the neutral, trusted platform where enterprises can safely deploy whichever models they choose, on their own data. The model layer is becoming a global battleground; the data platforms aim to be the stable ground beneath it.