U.S. chip export regulations scarcely hinder China’s A.I. industry. Last year’s U.S. semiconductor export limits to halt China’s supercomputer development for nuclear weapons and artificial intelligence systems like ChatGPT have no impact on China’s tech industry.

The limitations banned shipments of Nvidia Corp (NVDA.O) and Advanced Micro Devices Inc (AMD.O) processors, the worldwide technology industry’s standard for chatbots and other A.I. systems.

Nvidia has delayed its Chinese chips to comply with U.S. regulations. According to industry analysts, tAs a result, the Nvidia H800, launched in March, may take 10% to 30% slower to perform some A.I. tasks and quadruple some prices compared to Nvidia’s fastest U.S. processors.

Even sluggish Nvidia processors help Chinese enterprises. In April, Tencent Holdings (0700. H.K.), one of China’s leading tech businesses, predicted that Nvidia’s H800 would decrease the time it takes to train its largest A.I. system from 11 days to four days.

“The A.I. companies that we talk to seem to see the handicap as relatively small and manageable,” said Shanghai-based 86Research analyst Charlie Chai.

The government-industry conflict highlights the U.S. problem of restraining China’s high-tech advancement without damaging U.S. enterprises.

The U.S. created the limits to avoid a shock that might cause the Chinese to abandon U.S. chips and redouble their chip-development efforts.

“They had to draw the line somewhere, and wherever they drew it, they were going to run into the challenge of how to not be immediately disruptive, but how to also over time degrade China’s capability,” said one chip industry executive who requested anonymity to discuss regulator discussions.

Export limitations are dual. The first limits military supercomputers’ capacity to calculate exceedingly exact figures. Chip industry sources stated that it worked.

In A.I. work like large language models, chip capacity is more important than precision.

For such work, Nvidia offers the H800 to China’s biggest tech companies, including Tencent, Alibaba Group Holding Ltd (9988. H.K.), and Baidu Inc (9888. H.K.).

“The government isn’t seeking to harm competition or U.S. industry, and allows U.S. firms to supply products for commercial activities, such as providing cloud services for consumers,” Nvidia said last week.

It stated that U.S. technology sells well in China.

“The October export controls require that we create products with an expanding gap between the two markets,” Nvidia stated last week. We follow the law and offer competitive products in each market.”

In a second statement this week, Nvidia’s chief scientist Bill Dally claimed: “this gap will grow quickly over time as training requirements continue to double every six to 12 months.”

The U.S. Commerce Department’s Bureau of Industry and Security enforces the standards but does not respond.

The second U.S. barrier affects A.I.: chip-to-chip transmission rates. ChatGPT models are too huge for a chip. So instead, they must be distributed among numerous chips—often thousands—that must communicate.

The China-only H800 processor’s chip-to-chip throughput is 400 gigabytes per second, less than half the peak speed of Nvidia’s flagship H100 chip, accessible outside China.

A.I. experts think that’s fast. For example, MosaicML CEO Naveen Rao projected a 10-30% system delay.

“There are ways to get around all this algorithmically,” he remarked. “I don’t see this being a boundary for a very long time – like 10 years.”

Money aids. China’s slower chip can still teach A.I.

“At that point, you’ve got to spend $20 million instead of $10 million to train it,” one industry insider said, requesting anonymity due to partner agreements. That stinks. It does. Is this difficult for Alibaba or Baidu? No issue.”

A.I. researchers also strive to shrink their large systems to lower the cost of training products like ChatGPT. Fewer chips will reduce chip-to-chip connectivity and the influence of U.S. speed limitations.

Cade Daniel, a software developer at Anyscale, a San Francisco firm that helps organizations use A.I., said two years ago, the industry thought A.I. models would grow.

“If that were still true today, this export restriction would have a lot more impact,” Daniel added. “This export restriction is noticeable, but it’s not quite as devastating as it could have been.”