No Chips For You! Senator Wants Americans To Get First Dibs On GPUs, Restrict Sales To Others

+Comment US lawmakers are looking to apply Trump's America-First agenda to advanced semiconductors by giving US buyers first dibs while restricting the sale of most high-end chips needed for AI to the rest of the world.

"Artificial Intelligence is a transformative technology and United States policy should ensure that United States persons, including small businesses, startups, and universities are in the best position to innovate and harness the potential of artificial intelligence," the "Guaranteeing Access and Innovation for National Artificial Intelligence" or GAIN AI Act of 2025 reads.

If approved by Congress and signed into law by the president, the legislation, introduced as an amendment to the National Defense Authorization Act by Senator Jim Banks (R-IN), would require exporters seeking licenses to countries of concern to certify US buyers had the right-of-first-refusal on advanced silicon. It would also make vendors meet all domestic demand before they can flog their wares to foreign customers.

The rules would enact new trade restrictions mandating that exporters obtain approval from the US Commerce Department for any shipment of advanced semiconductors exceeding steep performance caps, which we've covered in depth previously.

"It should be the policy of the United States and the Department of Commerce to deny licenses for the export of the most powerful AI chips including such chips with total processing power of 4,800 or above and to restrict the export of advanced artificial intelligence chips to foreign entities so long as United States entities are waiting and unable to acquire those same chips," the proposed legislation reads.

These performance caps, which were enacted in large part during the Biden administration, were designed to prevent the United States' most powerful chips from reaching countries of concern, most notably China.

As these rules are written, most high-end datacenter GPUs and some gaming cards, like Nvidia's RTX 5090, would be subject to license restrictions.

Unsurprisingly, Nvidia isn't a fan of the proposed rules, which we'll emphasize again have not been signed into law.

"The AI Diffusion Rule was a self-defeating policy, based on doomer science fiction, and should not be revived. Our sales to customers worldwide do not deprive U.S. customers of anything — and in fact expand the market for many U.S. businesses and industries," an Nvidia spokesperson told El Reg. "The pundits feeding fake news to Congress about chip supply are attempting to overturn President Trump's AI Action Plan and surrender America's chance to lead in AI and computing worldwide."

El Reg Comment

To Nvidia's point, the GAIN AI Act makes several assumptions in support of the proposed rules, many of which are predicated on flawed logic and a fundamental misunderstanding of semiconductor and AI supply chains.

The bill makes the case that demand for advanced artificial intelligence chips far exceeds the supply, and US companies are being forced to wait months in order to get their hands on them.

While it's true that AI infrastructure is very much a seller's market at the moment, this is kind of how a free market is supposed to work. You snooze you lose.

However, the bill also misses a far more important element. The AI infrastructure supply chains aren't as simple as buying chips, or more specifically, GPU servers. They aren't much good if you don't have somewhere to put them. These systems can't be dropped in any old datacenter either. 

A single server can consume as much power as an entire rack from just a few short years ago. These systems often require advanced thermal management systems like direct to chip liquid cooling, rear-door heat exchangers, and coolant distribution units, all of which have to be in place before a single system can power on.

Nvidia's CEO has made it abundantly clear that the company prioritizes customers who have the datacenter capacity required to field the chips, so they're not sitting unused in a warehouse, while another customer that does have capacity ready to go is forced to wait.

The GAIN AI Act also calls out the production of advanced semiconductors specifically for sale in countries of concern — a rather unambiguous reference to Nvidia and AMD's China-spec H20 and MI308 accelerators, which were cut down to comply with performance caps enacted during the Biden administration.

The Trump administration cleared the way to resume shipments of these chips earlier this summer on the condition that Uncle Sam gets a 15 percent cut of all sales to China. 

The amendment also argues that more chips for China means fewer for American buyers. However, this overlooks just how uncompetitive these parts are compared to chipmakers' mainstream offerings.  

The H20 is based on now three year old technology and delivers an order of magnitude less performance (296 teraFLOPS FP8 vs. 5-15 petaFLOPS FP8/FP4) than the Blackwell-based alternatives currently available to US customers. And because these chips are based on an older process tech from TSMC, they don't even eat into production capacity that could go towards more attractive Blackwell parts as the bill suggests.

As such, the bill would only serve to limit Nvidia and AMD's addressable market while simultaneously creating new markets for increasingly competitive Chinese silicon to find a foothold. ®

RECENT NEWS

From Chip War To Cloud War: The Next Frontier In Global Tech Competition

The global chip war, characterized by intense competition among nations and corporations for supremacy in semiconductor ... Read more

The High Stakes Of Tech Regulation: Security Risks And Market Dynamics

The influence of tech giants in the global economy continues to grow, raising crucial questions about how to balance sec... Read more

The Tyranny Of Instagram Interiors: Why It's Time To Break Free From Algorithm-Driven Aesthetics

Instagram has become a dominant force in shaping interior design trends, offering a seemingly endless stream of inspirat... Read more

The Data Crunch In AI: Strategies For Sustainability

Exploring solutions to the imminent exhaustion of internet data for AI training.As the artificial intelligence (AI) indu... Read more

Google Abandons Four-Year Effort To Remove Cookies From Chrome Browser

After four years of dedicated effort, Google has decided to abandon its plan to remove third-party cookies from its Chro... Read more

LinkedIn Embraces AI And Gamification To Drive User Engagement And Revenue

In an effort to tackle slowing revenue growth and enhance user engagement, LinkedIn is turning to artificial intelligenc... Read more