Symbolic imageAnthropic and OpenAI break with US tech industry over Chinese open-weight AI models
Anthropic and OpenAI are at odds with much of the rest of the American technology industry over whether so-called open-source artificial intelligence models from China should circulate freely in the United States or be restricted, the New York Times reported on 25 July 2026. The dispute follows endorsements of open-weight models by Nvidia, Microsoft and IBM, reported by The National. At stake is how far Chinese systems spread through Western infrastructure.
What happened
According to the New York Times, Anthropic and OpenAI are clashing with the rest of the technology industry over whether "open-source" models from China should be freely available or restricted. The paper describes the disagreement as a split within Silicon Valley rather than a dispute between a company and a regulator: the question is whether the model weights of Chinese systems can be downloaded, run and built upon inside the United States without limits, or whether access to them should be curtailed. The reporting available here does not name a specific Chinese model, a specific proposed restriction, a government decision, a deadline or any figures, and no legislative or regulatory step is described as having been taken. The immediate background, as summarised in the topic context and attributed to The National, is that Nvidia, Microsoft and IBM are among the companies endorsing open-weight artificial intelligence models — that is, models whose parameters are distributed widely rather than held closed. That places three of the industry's largest hardware and software vendors on the side of broader distribution, and it is against this backdrop that the position of the two leading developers of closed frontier models stands out.
The camps
The dispute as documented runs between companies, not states. On one side, the New York Times places Anthropic and OpenAI, which are described as favouring restrictions on the free availability of Chinese open-source models in the United States. Both build models whose weights are not published, so their commercial approach coincides with the restrictive position, a point the available reporting notes only implicitly. On the other side stands what the New York Times calls the rest of the tech industry. The nearest concrete evidence for that side is the endorsement of open-weight models by Nvidia, Microsoft and IBM reported by The National — an endorsement of the open-weight principle in general, which is not reported there as a statement specifically about Chinese systems. The link between the two reports should therefore be treated as adjacency rather than as a confirmed alignment on the China question. No Chinese company, no Chinese authority and no United States government body speaks in the material available, so the side whose models are the object of the argument is not represented here.
The view from outside
The one non-American perspective in the material is that of The National in the United Arab Emirates, which frames the matter as a question of industry preference — which companies favour parameters being distributed more widely rather than held closed — rather than as a security or geopolitical contest. Beyond that framing, no Chinese, Global South or European coverage is present in the sources, so how the argument is read outside the United States cannot be established here.
What's new
The topic to date consisted of a single reported development: the endorsement of open-weight models by Nvidia, Microsoft and IBM. What is new is the emergence of a named opposing side. Where the earlier reporting showed only a direction of travel within the industry, the New York Times now describes an open split, with Anthropic and OpenAI set against the rest of the sector, and it sharpens the subject from open weights in general to Chinese open-weight models in particular. No new decision, rule or product accompanies this shift in the material available.
What could happen next
One path is that the argument stays inside the industry and is settled commercially: cloud providers, chipmakers and enterprise vendors continue to distribute and support Chinese open-weight models, and the restrictive position remains a minority stance among firms whose own models are closed. A second is that the dispute moves into policy, with the positions now visible in the press becoming submissions to United States regulators or lawmakers, in which case the availability of Chinese weights would be decided outside the companies. A third is that the split hardens into two operating environments, with parts of Western infrastructure standardising on open weights of any origin while others limit themselves to models from domestic closed providers. Which of these develops cannot be judged from the reporting at hand.