Public arguments about artificial intelligence often appear to be about principle. One executive defends open models. Another warns about national security. A third argues that regulation will concentrate power. But behind these statements sits a more practical question: which part of the technology industry should bear the cost of slowing China’s AI progress?
That question has moved to the centre of a growing dispute involving Anthropic CEO Dario Amodei, Nvidia CEO Jensen Huang, Meta CEO Mark Zuckerberg and former U.S. AI policy official David Sacks. What looks like a debate about open-weight models is increasingly becoming a conflict between business models, geopolitical priorities and competing visions of American AI leadership.
Anthropic Rejects Calls to Ban Open-Weight Models
Amodei recently published a statement clarifying that Anthropic does not support banning open-weight AI models. These are systems whose model weights are released publicly, allowing companies, researchers and governments to download, modify and run them independently.
The clarification followed criticism from Sacks, who argued that Anthropic had isolated itself by refusing to sign an industry letter supporting open AI models. Nvidia’s Huang had promoted the letter publicly, while Sacks said Anthropic should be watched “like a hawk.”
Amodei rejected the suggestion that Anthropic wanted to eliminate open competition.
“A ban would protect U.S. AI companies from competition, but that has never been my goal.”
His argument is that open models are not the central national security threat. The greater danger, in his view, is that an authoritarian state could develop AI systems more capable than those available in the United States and use them for military dominance or domestic repression.
The best way to prevent that outcome, he argues, is not to restrict access to Chinese models after they have been built. It is to restrict China’s access to the computing infrastructure required to build them in the first place.
That Position Places Nvidia at the Centre of the Conflict
Amodei’s reasoning shifts the burden of containment away from software companies and toward the semiconductor industry.
The implication is clear: if the United States wants to slow Chinese AI development, it should impose stronger restrictions on the export of advanced chips and related infrastructure.
That position conflicts directly with Nvidia’s commercial interests. The company has spent years attempting to preserve access to the Chinese market, which Huang has described as a potential tens-of-billions-of-dollars opportunity.
Restrictions on advanced Nvidia chips were eased under the Trump administration after earlier controls had limited sales into China. Amodei’s argument effectively suggests that those restrictions should have remained in place or become even stronger.
The dispute therefore raises a difficult policy question. Should American chipmakers sacrifice major international markets in order to protect the competitive position of U.S. AI labs?
Open Models Have Already Become Too Important to Contain
Even critics of Chinese AI recognise that open-weight models are becoming deeply embedded in the global technology market.
Models from companies such as Alibaba, DeepSeek and Moonshot AI are increasingly being used by startups and enterprises because they are cheaper, customizable and capable of running on private infrastructure. For many organizations, this is not simply an ideological preference for open source. It is a practical attempt to avoid dependence on a small number of American AI providers.
As Uljan Sharka, CEO of AI company Domyn, argued:
“Open source has to win. It will win. It’s inevitable.”
The economic appeal is clear. Open models reduce the cost of building software, launching businesses and automating internal processes. They allow companies to retain control over their data and adapt models to specific industries or national requirements.
A U.S. ban would therefore risk harming American startups and smaller businesses while doing little to prevent these systems from gaining users elsewhere in the world. Once capable open models are widely distributed, there is no realistic way to remove them from the market.
Chips Matter, but They Are No Longer the Entire Story
Amodei’s focus on semiconductors is based partly on the idea that greater computing power produces more capable AI systems. This principle remains broadly true and has shaped much of the industry’s investment in ever-larger clusters of advanced chips.
However, computing power is no longer the only source of AI progress.
Chinese companies have continued to make major advances despite export restrictions. They have compensated through improved algorithms, more efficient model architectures and techniques such as distillation, where smaller models learn from the outputs of larger systems.
This means export controls may slow development, but they are unlikely to stop it entirely. China also appears to be pursuing a different definition of AI leadership. While U.S. companies remain focused on building the world’s most capable frontier model, Chinese strategy increasingly emphasizes widespread adoption, low cost and broad distribution.
Under that model, a system does not need to be the smartest AI in the world. It only needs to be capable enough, cheap enough and widely available enough to become part of the global technology infrastructure.
The Open Model Debate Is Becoming a Commercial Battle
For years, the debate over open AI centred largely on safety. Supporters of closed systems argued that proprietary access made it easier to enforce safeguards and prevent malicious use.
That debate has not disappeared, but it is now increasingly mixed with commercial interests.
Closed AI companies benefit when customers must access their models through paid platforms and APIs. Open-model developers benefit when software can be widely downloaded, customized and distributed. Chip companies benefit when global demand for computing infrastructure remains unrestricted.
Each group can therefore present its own business interest as a national interest. Anthropic can argue that tighter chip controls protect American security. Nvidia can argue that open models and international market access support innovation. Meta can warn that restrictions on open systems will concentrate power in a small group of closed providers.
All of these arguments may contain genuine policy concerns, but they also protect the commercial position of the companies making them.
Regulation Could Reshape the Entire AI Market
The policy choices now being considered in Washington could determine which parts of the AI ecosystem continue to grow.
Restricting Chinese models could protect American AI labs domestically but make open Chinese systems more attractive in international markets. Tightening chip controls could slow Chinese development while reducing revenue for U.S. semiconductor companies. Stronger regulation of frontier models could improve safety but slow American labs as Chinese competitors continue to advance.
There is no option without economic consequences. The challenge for policymakers is to separate genuine national security concerns from attempts by powerful companies to shape regulation in their own favour. This will become increasingly difficult as the success of the stock market, the semiconductor industry and the AI sector become more closely connected.
The Deeper Strategic Question
The dispute between Amodei, Huang and Zuckerberg reveals that Silicon Valley may broadly agree on the importance of maintaining American AI leadership, but it does not agree on who should make the sacrifice.
AI labs do not want restrictions that protect open competitors. Open-model companies do not want regulation that strengthens closed providers. Chipmakers do not want to surrender international markets in order to support software companies.
Each side would prefer another part of the industry to absorb the cost of containing China. That may be the real meaning behind the public debate. The question is no longer simply whether AI should be open or closed, or whether China should be restricted.
It is whether the United States can build a coherent AI strategy when its most powerful technology companies have fundamentally different economic interests.
The future of AI policy may depend less on choosing between openness and safety and more on deciding which companies Washington is willing to inconvenience in the name of national security.
