In a happy development, the US government has pulled the plug on a proposal that was meant to regulate the export of artificial intelligence chips around the world.
The law, proposed by the US Department of Commerce, was published on a government website earlier this week, catching many in Silicon Valley off guard. It’s no secret that chips are an important part of AI models, and without them, many would not work.
The original document announcing the news is available on the agency’s rulemaking website here, and information about the withdrawal was first reported in a news article about the policy reversal here. But to reverse it, the proposed rule would have prevented the export of certain AI chips used in training models and data centers.
A draft of the proposal was circulated among government agencies in January, and it would remove the old framework that divided countries into groups based on how close they are to the US.
The news comes at a time when the fate of AI hardware is already fraught with politics. The US has banned the export of its most advanced chips to other countries, including China. Apart from this, companies have found ways to expand the power of AI computing overseas.
For example, Chinese Internet giant ByteDance will reportedly have access to a 36,000-GPU cluster in Malaysia, built using Nvidia Blackwell GPUs, as reported in this article about navigating US export restrictions.
All of this raises a question that lawmakers must answer: whether export restrictions hinder the development of AI systems in other countries or simply encourage them to find ways around the rules. Some argue that it is the latter. After all, it takes more than chips to build AI models: There’s software and other types of hardware involved.
Lawmakers have considered other ways to limit the export of semiconductor technology. In one case, foreign governments or companies may have to invest in US data centers, or commit to certain security agreements, in order to get permission to import a large number of AI chips.
The purpose of such restrictions, described in this article on proposed AI chip export rules, is to ensure that the US remains an AI hub while restricting sensitive technology from flowing to adversaries.
The global semiconductor industry, for one, has been closely following these negotiations. Chip companies, including Nvidia, AMD, and Intel, rely heavily on foreign markets, and any changes in export regulations can have far-reaching effects on supply chains, prices, and investment decisions.
Analysts say simple uncertainty surrounding the potential changes could have the effect of deterring investors, especially since AI data centers are expected to cost billions of dollars to build.
Complicating matters is that AI is a rapidly growing field, with demand for advanced chips as technology companies scramble to train larger AI models.
AI data centers are growing at a furious pace, and governments are increasingly viewing AI computing capabilities as a valuable resource, similar to energy or military assets. In this way, the battle for the future of AI could be a battle for who controls the chips. If it sounds overwhelming, it is.
Here’s the thing: no one is sure yet what any export policy will ultimately look like. Lawmakers are trying to balance national security concerns with the desire to maintain the competitiveness of US technology companies.
Sometimes those goals are different. As negotiations continue in Washington, industry watchers expect that the US will eventually issue new proposals, perhaps with stricter restrictions, or perhaps with additional agreements.
In any case, the law is dead for now, the debate continues, and the race to build AI systems around the world continues unabated. Perhaps today’s news will be a footnote in history, or perhaps it will serve as the beginning of a much larger movement.
