NVIDIA H20 Sales Relief Won’t Hinder Chinese Firms from Developing AI Chips “In Parallel,” According to Research Firm

Please note, this article does not constitute investment advice, and the author holds no positions in the stocks discussed.

Following NVIDIA’s announcement regarding anticipated export licenses for H20 GPU sales to China, a new report from TrendForce suggests that local chip designers and semiconductor manufacturers in China will continue their development endeavors. Earlier this year, the U. S.imposed restrictions on H20 exports, prompting NVIDIA CEO Jensen Huang to assert that such limitations would inadvertently accelerate the growth of China’s domestic chip ecosystem. According to TrendForce, China’s procurement ratio for foreign AI chips is projected to rise to 49%, an increase of 7 percentage points from the previous estimate of 42%.

Rising Demand for NVIDIA’s H20 Chip in China

TrendForce corroborates prevailing views that, despite stringent measures, Chinese entities like Huawei have struggled to keep pace with NVIDIA’s technological advancements. A primary obstacle for Chinese companies in developing competitive AI chips is their reliance on outdated manufacturing technologies, stemming from U. S.restrictions that prevent access to cutting-edge lithography equipment. Additionally, Huawei’s inability to procure advanced chips from TSMC in Taiwan further confines them to older technologies for chip production.

The anticipated resumption of H20 shipments is expected to unleash a substantial demand for AI infrastructure in China, particularly among major Cloud Service Providers (CSPs).The introduction of high-bandwidth memory (HBM) is also a significant factor driving interest in the H20 chips. The 2024 H20 chips will employ HBM3 8hi memory, with plans for an enhanced HBM 3e 8hi in 2025, poised to attract considerable attention as domestic firms navigate existing export restrictions.

NVIDIA H20 GPU Image — Image Credit: servethehome.com

In his discussions with the Trump administration about resuming sales of the H20 GPUs to China, Huang emphasized that allowing these sales could position American technology as a foundation for global AI hardware. He likened the GPUs to the U. S.dollar, suggesting they could become a worldwide standard in AI computing.

It’s worth noting that NVIDIA’s H20 GPUs face strict design constraints that reduce their operational efficiency to comply with U. S.export regulations. These GPUs are based on the older Hopper architecture, while competitors from non-sanctioned countries have access to products built on the more advanced Blackwell architecture.

Huang has attempted to address U. S.national security concerns by highlighting that the Chinese military is unlikely to depend on foreign chips for its computing requirements. He posited that any stringent restrictions would only expedite the efforts of local chip designers to create alternative solutions. However, TrendForce warns that the H20’s return to the Chinese market may lead local manufacturers to aggressively stockpile NVIDIA products, while the AI sector in China remains susceptible to geopolitical tensions, leaving NVIDIA at risk.

Ultimately, TrendForce concludes that as Chinese Original Equipment Manufacturers (OEMs) and CSPs ramp up their stockpiling of NVIDIA technology, domestic AI chipmakers—bolstered by governmental backing—will concurrently accelerate their development efforts.