Nvidia’s chip allocation power is becoming political power

OpenAI has continued to support export controls in general, but it has avoided taking a detailed public position on advanced chip sales to China. The Information reports that leaders including Sam Altman worried a stronger stance could alienate Jensen Huang and Nvidia. A recent internal question from researcher Noam Brown—asking what the company’s policy actually was—received references to older principles and an invitation to discuss the developing position privately rather than a clear answer in Slack.

The leverage begins with allocation. Nvidia’s GPUs remain the dominant input for training and serving frontier models. Cloud executives told The Information that the risk of angering the company sits in the background of purchasing and marketing decisions; some customers did not want a provider to advertise that it also operated AMD chips. Other buyers have felt that Nvidia favoured customers purchasing its complete hardware stack rather than components à la carte. No formal retaliation has to be demonstrated for scarcity to discipline behaviour: customers know that their next cluster depends on a continuing commercial relationship.

The same tension runs through Washington. Nvidia argues that loosening some China restrictions helps keep the global AI stack on American hardware. Critics say the company exaggerates the competitiveness of Chinese alternatives to strengthen that case. Huang’s access to President Trump has produced limited permission for sales of older chips, while the company’s lobbying has also created distrust among bipartisan China hawks. Reports that Nvidia pressed think tanks over positions it disliked make the influence problem broader than ordinary lobbying: the firm funds parts of the policy ecosystem evaluating its preferred policy.

Nvidia is simultaneously tightening the overseas-compute route that has weakened export controls. It has told some data-centre partners to make at least 51% of future cloud business non-Chinese, and has asked providers to identify the ultimate users of its chips. Chinese developers can rent restricted hardware abroad even when they cannot import it, so customer-composition rules turn Nvidia’s private sales process into an enforcement layer.

That contradiction is the important one. Nvidia campaigns against restrictions that limit its addressable market, then imposes its own controls to preserve political permission to sell globally. Its influence does not come only from size or access to the president. It comes from occupying the point where model companies, cloud providers, exporters and regulators all need the same scarce product—and can therefore be steered through allocation decisions that never appear in legislation.