Media & Culture

China snubs Nvidia H200 chips, spurning $30B market

China refuses Nvidia H200 chips despite US approvals, costing $30B...

Deep Dive

Nvidia’s H200 AI chips—recently granted export approval to Chinese firms like Alibaba and Tencent by the US Commerce Department—have hit a brick wall in China, where the government has refused to authorize any purchases. This diplomatic and commercial setback represents a stunning reversal for Nvidia CEO Jensen Huang and former President Donald Trump, who personally lobbied Chinese President Xi Jinping to greenlight the sales during a high-profile meeting. The move comes amid escalating tensions over chip access and AI sovereignty, with Beijing increasingly prioritizing domestic chipmakers such as Huawei’s Ascend series over U.S. alternatives.

The refusal isn’t just a strategic pivot—it’s a calculated economic and security decision. Analysts point to a 25% tariff imposed by the U.S. on Nvidia’s China-bound chips, which significantly inflates costs for Chinese buyers. Additionally, Beijing has raised serious concerns about the risk of foreign tampering during transit, fearing potential spyware or backdoor risks when chips pass through U.S. territory. As a result, Chinese AI developers are doubling down on domestic hardware, accelerating the optimization of models like DeepSeek V4 for Huawei’s Ascend chips. For global users, this could mean future compatibility gaps, particularly for AI workloads optimized for Chinese hardware ecosystems.

Key Points
  • China blocked all H200 AI chip purchases from Nvidia despite US export approvals, potentially costing Nvidia $30B in lost sales
  • Beijing cites cost (25% tariff) and security fears over chip tampering during transit as reasons for the rejection
  • Chinese AI models like DeepSeek are increasingly optimized for domestic hardware (e.g., Huawei Ascend), reducing reliance on Nvidia

Why It Matters

The move accelerates China’s AI self-sufficiency, reshaping global AI hardware supply chains and leaving Nvidia’s dominance in flux.