U.S. Tries to Crush China's AI Ambitions With Chip Crackdown

"New U.S. chip-export limits that rocked global markets Wednesday are the clearest sign yet from the Trump White House that whatever advances China makes in AI will have to happen without the U.S.'s help.

Trump administration officials have signaled for months that they were considering a crackdown on exports of processors from U.S. companies such as Nvidia that have helped enable major Chinese advances in artificial intelligence. The latest reckoning came this week, with U.S. authorities moving to stop the flow of billions of dollars worth of Nvidia and Advanced Micro Devices artificial intelligence chips to the country.

The move, spurred in part by Chinese AI startup DeepSeek's surprising success in building powerful models with less computing power, put an end to Nvidia's ability to skirt U.S. constraints on sales by tweaking its chips. While the changes impact a relatively small portion of the companies' business, they squash any hopes of unimpeded future chip sales to China.

Nvidia shares tumbled nearly 7% Wednesday, along with fellow chip company AMD. The broader stock market also sagged on news that any hope of a reprieve from the highest "reciprocal tariffs" imposed on Chinese exports was short-lived.

The new restrictions, which impact Nvidia's H20 chips and AMD's MI308 products, capped a roughly two-week blitz that created market whiplash and underscored the administration's determination to engage China on many fronts. The battle for tech supremacy is set to unfold in much the same way as the trade war: with toughness, and a bit of chaos.

Trump imposed a 145% tariff last week, and then exempted processors, smartphones and other electronics from it days later. On April 9, the federal government told Nvidia that it would be subject to new export restrictions. The following Monday, Nvidia said it would build AI supercomputers in Texas, in line with the president's desire to bring semiconductor manufacturing back to American shores. The next day, it disclosed the new restrictions.

"Even when you think you're playing by the rules and sacrificing a bit for the team, the rules might change tomorrow and you're going to get hit again," said Scott Lincicome, vice president of general economics and trade policy at the Cato Institute, the libertarian think tank. "It's a brutal environment for folks who want to spend billions of dollars on a 30-year project."

Trump administration officials have increasingly felt that new innovations in AI were driven by inference [1], which allows AI models to be applied in real-life scenarios.

Nvidia's H20 had proved to be very effective in that work, people familiar with the matter said.

In recent weeks, U.S. officials, including from the National Security Council, recently met with Nvidia chief Jensen Huang to ask about the H20 and the company's global supply chains, the people said.

The administration is facing a deadline in May on how to handle the sales of advanced AI chips globally, another decision that will shape the industry's future. Meanwhile, it is investigating the sector and other tech products containing semiconductors for national-security vulnerabilities as part of its tariff strategy.

Publicly, Nvidia has said selling to China helps bring in revenue that it uses to keep its global lead in AI.

Privately, it has pushed back against any new restrictions, arguing that China has already been able to produce some chips comparable to its H20, people familiar with the issue said.

In the fiscal year ended January, Nvidia sold around $12 billion worth of H20s, accounting for about 70% of the company's China sales, according to analysts.

China's ability to access Nvidia's H20s had been on the radar of U.S. policymakers since the last administration. Officials who realized the value of H20s in AI development had privately discussed how to handle purchases of the chip, The Wall Street Journal has reported.

Such talks continued under the Trump administration, and were given added urgency earlier this year when Chinese AI startup DeepSeek surprised Silicon Valley and Wall Street with its low-cost, but powerful models. The emergence of DeepSeek has boosted the adoption of AI models in China, driving up demand for Nvidia's H20 and other chips.

During his nomination hearing in January, Commerce Secretary Howard Lutnick said Nvidia chips drove the development of DeepSeek's model and such U.S. assistance had to end.

"If they're going to compete with us, let them compete, but stop using our tools to compete with us," he said. "I'm going to be very strong on that."

On Wednesday, a new report on DeepSeek from a congressional committee focused on China recommended tightening export controls and boosting funding for the Commerce Department office that oversees them.

Chinese customers rushed to make purchases, anticipating a ban would come soon. In the first three months of this year, Nvidia received around $18 billion worth of orders for servers and modules equipped with the H20, a higher value of orders than Nvidia's total revenue from China during the last fiscal year, according to people familiar with the purchases. The biggest buyers were Chinese cloud computing companies such as Alibaba, Tencent and ByteDance, the people said.

If the U.S. doesn't grant any exemptions, the move would cut off an important source of computing power used by Chinese companies and research houses, and could increase demand for local substitutes.

Chinese cloud-service providers originally planned to source 50% of AI accelerator demand in 2025 using H20 processors and would now likely turn to domestic suppliers Huawei and Cambricon for AI chips, analysts from Citi wrote in a report Wednesday.

Beijing has been pushing for the development of its own chip industry independent of U.S. technology and encouraged wider adoption of domestic substitutes." [2]

 

1. A good inference is a conclusion reached by combining evidence and reasoning, going beyond what is explicitly stated, and using prior knowledge and textual clues to understand deeper meanings or implications.

    Definition:

    An inference is an educated guess or a conclusion drawn from available evidence and reasoning, rather than being directly stated.

Reading Comprehension:

In reading, inferences help readers understand what is implied or not directly stated in the text.

Textual Clues:

Good inferences rely on identifying clues and details within the text, along with the reader's background knowledge.

Prior Knowledge:

Readers use their existing knowledge and experiences to interpret the text and make informed conclusions.

Examples:

 

    Example 1: If a character is described as "sweating and looking pale" after a scary encounter, you can infer they are frightened.

 

Example 2: If a text mentions "a family gathering" and then "a plate of cookies", you can infer they are celebrating a special occasion.

 

Key Steps in Making Inferences:

 

    Read the text carefully: Pay attention to details and clues.

 

Identify clues and patterns: Look for words, phrases, and events that might suggest something not directly stated.

Consider your background knowledge: Think about what you already know that might help you understand the text.

Draw a conclusion: Formulate a logical inference based on the clues and your knowledge.

 

2.  U.S. Tries to Crush China's AI Ambitions With Chip Crackdown. Lin, Liza; Ramkumar, Amrith. Wall Street Journal, Eastern edition; New York, N.Y.. 17 Apr 2025: A1.