Skip to content

Beijing authorities advocating caution towards NVIDIA's H20 AI GPUs due to trust concerns

NVIDIA faces renewed challenges in its efforts to sell H20 AI GPUs to China, as reports indicate Beijing is encouraging local companies to shun the chips.

China Expresses Skepticism Towards NVIDIA - Trust Concerns Surround H20 AI GPUs as Chinese...
China Expresses Skepticism Towards NVIDIA - Trust Concerns Surround H20 AI GPUs as Chinese Authorities Advise Against Use, Due to Rapid Increase in Suspicion

Beijing authorities advocating caution towards NVIDIA's H20 AI GPUs due to trust concerns

In the midst of escalating tensions between the U.S. and China over advanced AI technology, Chinese authorities have expressed concerns about NVIDIA’s H20 AI GPUs. These concerns are primarily driven by national security and political implications, as the H20, although a somewhat limited version of NVIDIA’s high-end AI chips, remains powerful enough to raise alarms about China's rapid AI development.

The context behind this concern includes U.S. export controls since 2022, aimed at preventing China from acquiring the most advanced AI accelerators like NVIDIA’s A100 and H100. In response to tightened U.S. restrictions, NVIDIA designed the H20 GPU specifically with reduced capabilities to comply with export rules. However, in April 2025, U.S. authorities abruptly banned even the H20 exports to China, signaling that any sufficiently capable AI hardware was politically sensitive.

China designated shipments of NVIDIA’s H20 GPUs as strategic assets and instructed domestic cloud providers to halt new purchases to examine the chips for potential security threats. This move indicates Beijing’s concern about possible vulnerabilities or espionage risks, as well as the desire to reduce dependence on U.S. technology and boost domestic alternatives.

The political implications include the technology competition and trade conflict between the U.S. and China: U.S. sanctions aim to slow China's AI advancement, while China aims for semiconductor self-sufficiency but still faces ecosystem and performance limitations in domestic chips. The security concerns on both sides center on the dual-use nature of AI chips: powerful AI accelerators can be used not only for commercial applications but also for military AI, surveillance, and other strategic purposes, making their export and use tightly controlled and politically sensitive.

NVIDIA, the leading supplier of AI GPUs worldwide, has been affected by these concerns. As part of the final deal to receive export licenses, NVIDIA and AMD agreed to pay 15% of revenue garnered from AI GPU sales to China, specifically from NVIDIA's H20 and AMD's MI308 chips. NVIDIA estimated a need for 300,000 more H20 AI chips from TSMC to meet China's demand, in addition to the 600,000 to 700,000 chips already stockpiled.

In an attempt to assuage the fears of the Chinese market, NVIDIA published a blog post titled "No Backdoors. No Kill Switches. No Spyware." However, the situation could get worse for NVIDIA before long, as Beijing is pressuring major firms to produce answers as to why they are purchasing NVIDIA’s H20 GPUs.

The AI GPU export ban was reversed on July 14, as announced in an NVIDIA blog post. President Trump called NVIDIA’s H20 chips "obsolete" during an August 11 press conference, but the reversal of the ban was a result of an agreement between NVIDIA CEO Jensen Huang and President Trump. The exact terms of the agreement remain undisclosed.

Read also:

Latest

Forged an alliance with The Partners

Partners Firm Bonds Reinforced by Action

Claudio Struck has assumed the role of Senior Advisor at The Partners consulting firm since January 2025, bolstering the company's offerings in sectors like public affairs, construction and real estate, and international cooperation due to his extensive experience. Prior to this, Struck served...