Arm warns US CPU export ban

Rene Haas used a simple comparison to describe the problem Washington faces if it tries to expand chip controls from high-end graphics processors to AI-capable central processing units. In an interview with Reuters on the sidelines of Computex in Taipei on June 2, the Arm chief executive said CPUs are so widely used across computing that a ban on exports to China would be much harder to define and enforce than existing limits on specialized AI accelerators. Reuters reported that Haas said regulators would struggle to set performance and memory thresholds for CPUs in the way they have for Nvidia’s graphics chips. That warning lands as the U.S. government is tightening another part of its semiconductor regime. Reuters reported that Washington has moved to stop advanced AI chips from reaching Chinese companies through subsidiaries outside China, an effort aimed at closing a loophole in earlier restrictions. CNBC, citing Commerce Department guidance, said the step followed concerns that Chinese firms had been able to buy restricted chips through overseas entities for almost a year. (money.usnews.com) ### Why did Haas say CPUs are harder to police than GPUs? Haas told Reuters on June 2 that banning artificial-intelligence CPUs would be “nearly impossible” because CPUs are general-purpose chips used across servers, cloud systems and other products, not only in AI training clusters. He said the U.S. could try to limit them, but the category was harder to control than AI chips already covered by export rules. (money.usnews.com) Reuters said Haas made the comments as Arm promoted its AGI CPU at Computex. Arm announced ByteDance and Oracle as new customers for that processor, which it unveiled in March, and Haas said demand was stronger than it had been eight weeks earlier. Arm last month doubled its guidance for demand for the chip to $2 billion across its 2027 and 2028 fiscal years and said it sees the product generating $15 billion in annual revenue in about five years. (money.usnews.com) ### What changed in U.S. enforcement? The Commerce Department has been trying to prevent Chinese access to advanced AI chips, but critics say enforcement lagged. The Foundation for Defense of Democracies published a June 1 policy brief saying Commerce had effectively acknowledged that earlier controls did not fully cover Chinese-owned entities operating abroad for more than a year. That assessment is from FDD, not the government, but it tracks with subsequent U.S. guidance aimed at Chinese firms outside China. (money.usnews.com) Late-May guidance described by CNBC said export restrictions apply to Chinese companies overseas as well as firms operating inside China. That change matters because many U.S. controls have focused on specific high-performance chips and named destinations, leaving companies and regulators to test the boundaries through affiliates, cloud access and third-country supply routes. (fdd.org) ### Why does the supply-chain backdrop matter? South Korea’s trade data show the regional semiconductor boom has continued even as Washington tightens controls. The Dong-A Ilbo reported that South Korea’s exports reached a record $87.75 billion in May, with chip exports at a record $37.2 billion, up 169.4% from a year earlier. The report said stronger AI investment by U.S. technology companies including Google and Meta helped drive demand. (cnbc.com) The same data showed how concentrated that growth has become. Semiconductors accounted for 42.3% of South Korean exports in May, up from 24% a year earlier, according to the Dong-A Ilbo report. That does not measure Chinese access to restricted U.S. chips directly, but it shows how AI demand is reshaping trade flows across allied manufacturing hubs while U.S. regulators revise the rules. (donga.com) ### What is the immediate next question for Washington? The next test is whether U.S. officials keep targeting narrow categories of advanced chips or try to widen controls to more general-purpose processors. Haas’s comments suggest industry executives expect that line-drawing exercise to get harder as AI workloads shift from training toward inference, where CPUs, GPUs and custom chips can all play a role. Reuters also reported that Intel and Advanced Micro Devices have seen rising demand tied to AI applications involving autonomous software agents. (donga.com) For now, the clearest near-term milestone is the Commerce Department’s implementation of its updated guidance on Chinese firms’ overseas access to advanced chips. Arm, meanwhile, is trying to secure enough wafer supply from Taiwan Semiconductor Manufacturing Co. and is working with Japan’s Socionext as demand for its AGI CPU rises, Haas told Reuters in Taipei on June 2. (money.usnews.com)