The Trump Administration likes to veto things. Now they seem to want to do it with the AI models of Chinese companies, which are becoming increasingly competitive. The launch of Kimi K3 seems to have been the trigger for this new plan to be activated, but there is a problem: vetoing those models is a terrible idea.
This comes from afar. The US Department of Commerce I had already studied last year included several Chinese AI startups, including DeepSeek, in its famous Entity List. With this they wanted to limit the access of these companies to sensitive hardware and technology developed in the US.
Companies at risk for using Chinese models. Recently it has even been proposed drafting an executive order to hold US companies responsible for security breaches that appear due to using Chinese models in their systems. The objective was always the same: to discourage the use of these Chinese models as much as possible.
Why US companies use Chinese models. The reason is simple: Chinese open weights like DeepSeek V4 or Kimi K3 allow companies to download and run them on their own servers, dramatically reducing inference costs and keeping all data private. Coinbase CEO Brian Armstrong himself has indicated that use models such as GLM-5.2 and Kimi K2.7 in local production, which has allowed them to cut their total spending on AI in half despite the fact that token consumption has skyrocketed.
Duopolies without competition. David Sacks, White House AI advisor, posted a message on X on Sunday in which he warned of the risk of using these models: “We are at a critical turning point in AI policies. The leading laboratories with proprietary models, which are already a duopoly in terms of revenue from their AI models, want the government to eliminate Open Source competition.” A Axios report reveals that indeed both OpenAI and Anthropic could have part of the responsibility in promoting this ban.
This could be a shot in the foot for the US.. An analysis published in The Washington Post raises an argument worth considering. Treating open models as a security threat is confusing competition with a danger that must be contained. This text recalls how the Sears chain was not allowed to ban Walmart, nor IBM to ban Compaq or Dell, nor traditional airlines to veto the operators that lowered prices. In each case the same thing happened: an established company ran into a rival that was lowering costs, so it had only two options: compete or lose.
Linux and Open Source have already shown the way. As the author of the article says, open source eliminated the barriers of commercial software, which locked users into an alternative from which they had no way out. That did not make these companies disappear, but rather boosted competition. Red Hat, MongoDB, Android or Kubernetes showed that “giving away” the product was not incompatible with building profitable businesses around that product.
Danger, duopoly. That analysis shows that almost all companies prefer a scenario in which open models remain available. The only ones who have a direct interest in maintaining closed models are precisely Anthropic and OpenAI, because their businesses depend precisely on there being no free (or very cheap) competitive alternatives.
If they are so good, why are they afraid? What’s ironic is that if Anthropic and OpenAI really claim to be so far ahead of Chinese AI companies, they shouldn’t have to worry about the competition. Nor would they have to ask the government for help to stop their competition. The US antitrust laws themselves exist precisely to prevent a market from falling into the hands of one or two companies. This veto would precisely allow them to create that monopoly (or duopoly) to lock users and companies into it.
GIPHY App Key not set. Please check settings