A public disagreement has erupted between chipmaking giant Nvidia and AI developer Anthropic concerning U.S. export controls designed to limit China’s access to advanced artificial intelligence hardware.
The conflict centers on the necessity and strictness of the “AI Diffusion Rule,” a framework established under the previous administration and now facing potential revisions by President Trump’s officials, according to recent reports. Anthropic advocates for stricter enforcement and tighter rules, citing national security and detailing alleged Chinese smuggling efforts, while Nvidia pushes back, arguing the restrictions harm U.S. innovation and dismissing some claims as unfounded.
Anthropic, an AI research company significantly backed by Amazon, laid out its case for robust export controls in an blog post today, responding to the Commerce Department’s AI Diffusion framework.
The company contends that “maintaining America’s compute advantage through export controls is essential for national security and economic prosperity” as AI models become more powerful.
Anthropic pointed to sophisticated Chinese efforts to circumvent existing rules, alleging chips were smuggled hidden in “prosthetic baby bumps” and “packed alongside live lobsters.” To address these risks, Anthropic recommended lowering the amount of high-performance computing hardware – measured by processing power like TFLOPS (trillions of floating-point operations per second) – that can be exported without a license to certain countries (Tier 2 nations).
They also urged increased funding for the Bureau of Industry and Security to bolster enforcement and suggested adjusting the tiered system, potentially allowing secure Tier 2 countries more access via government-to-government agreements. They stressed the urgency, warning against delaying the rule’s May 15, 2025, effective date to prevent further stockpiling by Chinese firms.
Nvidia, whose hardware is crucial for training large AI models like Anthropic’s, offered a sharp rebuttal. A company spokesperson told CNBC, “American firms should focus on innovation and rise to the challenge, rather than tell tall tales that large, heavy, and sensitive electronics are somehow smuggled in ‘baby bumps’ or ‘alongside live lobsters.'”
The chipmaker argues that overly restrictive policies undermine American competitiveness. “China, with half of the world’s AI researchers, has highly capable AI experts at every layer of the AI stack. America cannot manipulate regulators to capture victory in AI,” the spokesperson added.
This stance aligns with recent comments from Nvidia CEO Jensen Huang, who, speaking in Washington D.C. on April 30th, emphasized the need for U.S. companies to build domestic “AI factories” and touted the job creation potential, while also acknowledging China is “not behind” in AI development.
Policy Background: The AI Diffusion Rule and H20 Ban
The current friction follows the Trump administration’s decision, effective April 15, 2025, to require specific, difficult-to-obtain licenses for exporting Nvidia’s H20 and AMD’s MI308 chips to China.
This rule, justified by the Commerce Department as necessary “to safeguard our national and economic security,” effectively blocked sales of the H20. This chip was Nvidia’s attempt to comply with earlier U.S. export controls enacted under the Biden administration, designed with significantly reduced performance density (around 2.9 TFLOPS per square millimeter compared to the H100’s 19.4 TFLOPS/mm²) after more powerful chips like the A100, H100, and A800/H800 were barred from China.
Despite being a step down, the H20 ban forced Nvidia to take a substantial $5.5 billion charge against its quarterly revenue due to stranded inventory.
This tightening occurred despite conflicting signals just days before suggesting a possible pause, potentially linked to Nvidia investment promises or a meeting between Huang and President Trump. Political pressure had also mounted, with Senator Elizabeth Warren stating in an April 14th letter, “The Commerce Department cannot further delay undertaking necessary and urgent action on the H20 to protect U.S. national security.”
Nvidia’s initial opposition to the underlying AI Diffusion Rule framework dates back to its announcement in January 2025 under the Biden administration. At that time, Nvidia VP Ned Finkle criticized the framework as “unprecedented and misguided overreach,” contrasting it unfavorably with Trump-era policies and arguing it threatened global innovation.
Competitive Shifts and Huawei’s Rise
The U.S. export restrictions are visibly reshaping the competitive landscape within China. Following the H20 ban, Chinese tech giant Huawei announced its next-generation Ascend 920 AI accelerator and prepared for mass shipments of its Ascend 910C processor starting this month.
The 910C, reportedly using advanced chiplet techniques (combining smaller silicon dies) to integrate two previous-generation processors, is positioned as a key domestic alternative. Huawei also unveiled its AI CloudMatrix 384 cluster, which leverages 384 of these 910C chips and advanced optical networking.
While this cluster reportedly surpasses Nvidia’s flagship GB200 system in raw processing power (measured in PFLOPS, or quadrillions of floating-point operations per second) and memory capacity, analysis indicates it does so with significantly higher energy demands.
This aligns with analyst Patrick Moorhead’s prediction, reported by the NYT via Winbuzzer, that “Chinese companies are just going to switch to Huawei.” In response to the ban, Nvidia CEO Jensen Huang visited Beijing on April 17, meeting with trade officials and the founder of AI firm DeepSeek, reportedly stating Nvidia hoped “to continue to cooperate with China.”
Enforcement Hurdles and the Path Forward
The effectiveness of export controls hinges on enforcement, a point underscored by Anthropic’s smuggling concerns and reports surrounding Huawei’s supply chain.
While China’s SMIC foundry can produce 7nm-class chips, analysis cited by Reuters suggests Huawei’s current advanced chips may utilize restricted components, allegedly sourced through intermediaries like Sophgo (using TSMC fabrication) and CoAsia Electronics (for Samsung HBM memory – High Bandwidth Memory, crucial fast memory for AI).
This purported link has reportedly led to U.S. scrutiny of TSMC. Huawei has denied using TSMC-made Sophgo chips. These alleged activities highlight the difficulties in preventing controlled technology from reaching targeted entities, bolstering Anthropic’s call for stronger measures.
Simultaneously, the restrictions are widely seen as accelerating China’s push for semiconductor self-sufficiency, backed by massive state funding. As the May 15 effective date for the AI Diffusion Rule approaches and the Trump administration weighs potential changes, the debate between prioritizing national security through controls versus fostering global competitiveness continues to intensify within the AI industry.
