Why Arm’s CEO Says Blocking AI CPU Exports to China Is Nearly Impossible

Haas was direct about the potential scope of such a rule. He suggested the U.S. would effectively have to limit "everything," a scenario he characterized as a "pretty hardcore cut" that may be impractical to enforce . This difficulty stands in sharp contrast to GPU controls, which can more feasibly be pegged to specific processing power and bandwidth metrics tailored to AI workloads.

The Loophole That Just Closed: Policy Whiplash in Real Time

Haas’s comments did not occur in a vacuum. Just days before Computex, on May 31, 2026, the Commerce Department’s Bureau of Industry and Security (BIS) issued guidance closing a loophole that had allowed overseas subsidiaries of Chinese firms to purchase advanced chips like Nvidia’s Blackwell and AMD’s MI350x . The new rule mandates that any entity with a parent company headquartered in China or Macau must obtain a BIS license for advanced AI semiconductors, regardless of where the receiving subsidiary is based .

This action represents a tightening cycle after a period of conspicuous loosening. The Trump administration rescinded the Biden-era AI Diffusion Rule in May 2025, with Commerce officials arguing it would have stifled innovation . In December 2025, President Trump announced approvals for H200 chip exports to China, a decision that drew fire from congressional China hawks and was later formalized by BIS in a case-by-case licensing posture . The May 31 subsidiary loophole closure thus signals a partial re-tightening, creating an environment of regulatory whiplash that complicates corporate planning.

The Ecosystem Independence Argument

At a foundational level, Haas has consistently argued that export controls carry a significant strategic risk. Speaking in 2025 at the Founders Forum Global conference, he warned, "If you narrow access to technology and you force other ecosystems to grow up, it’s not good... It makes the pie smaller, if you will. And frankly, it’s not very good for consumers" . He aligned himself with Nvidia CEO Jensen Huang, who has similarly criticized the long-term impacts of supply-chain decoupling.

The reasoning is straightforward: by denying advanced CPUs to China, Washington incentivizes the development of fully independent rival semiconductor ecosystems. This not only shrinks addressable markets for companies like Arm — which still derives roughly 15% of its revenue from China — but also accelerates the global fragmentation of technology standards .

The New Reality: ByteDance and Oracle Join Arm's AGI CPU Ecosystem

Further complicating the geopolitical picture is the commercial reality on display at Computex. Haas announced that ByteDance and Oracle have adopted Arm’s new in-house AGI CPU, its first production-ready data-center processor designed specifically for "agentic AI" workloads . They join Meta, OpenAI, and others in what is rapidly becoming a mainstream Arm-based server ecosystem, positioning Arm's silicon as a structural alternative to Intel’s Xeon and AMD’s EPYC lines .

ByteDance's participation is particularly notable given that it is a China-headquartered company using cutting-edge Arm data-center CPUs. The BIS subsidiary rule would almost certainly apply to any advanced U.S.-origin chips ByteDance seeks to acquire, raising questions about how Arm’s own chip business navigates the same restrictions its CEO critiques.

The Unresolved Tension

Haas's Computex message reveals the core dilemma facing U.S. export controls. On one hand, the government is aggressively closing loopholes and extending licensing requirements to foreign subsidiaries. On the other, the CEO of one of the world’s most critical chip architecture firms argues that extending this logic to CPUs is operationally unworkable and strategically counterproductive.

CPUs are not a specialized tool for AI; they are the general-purpose silicon that powers everything. Regulating them as dual-use AI components threatens to collapse a distinction that the semiconductor industry — and the global economy — rely on. As Haas put it, the U.S. can try, but it’s a much harder area to control than dedicated AI chips, and the unintended consequences could reshape the industry in ways no one can fully predict .