Nvidia defies US export controls: AI chip sales in China are booming

Nvidia is on track to sell $12 billion worth of artificial intelligence (AI) chips in China this year, despite US export controls. Although the restrictions have complicated business in one of the world's largest semiconductor markets, the $3 trillion Silicon Valley company plans to deliver more than a million of its new H20 chips in the coming months. These chips are designed to circumvent US regulations that restrict the sale of AI processors to Chinese customers. According to analysts' forecasts, Nvidia will sell nearly twice as many chips as Huawei's competing product, the Ascend 910B. Nvidia is the latest example of a Silicon Valley company caught in the tensions between Washington and Beijing. The Biden administration aims to prevent the world's most powerful chips from reaching China, fearing they might be used there to develop more potent AI systems with military applications. The resulting shortage of AI chips is impairing the competitiveness of Chinese technology companies like ByteDance, Tencent, and Alibaba against US-based firms such as OpenAI, Microsoft, Meta, and Google. Each H20 chip costs between $12,000 and $13,000, indicating that Nvidia is expected to generate more than $12 billion in revenue. This would exceed the $10.3 billion in revenue generated from its entire China business in the fiscal year up to January 2024. Nvidia and Huawei declined to comment on the forecasts. Since the Biden administration introduced restrictions on the sale of Nvidia's most powerful AI chips to China in 2022, the US company has warned that its business would suffer. Cloud computing providers and AI startups would have to turn to local alternatives such as Huawei. Although sales in China were lower before the launch of the H20 chip in the spring, analysts from Morgan Stanley and SemiAnalysis report that the chip is now being delivered in large quantities and is popular among Chinese customers despite its relatively limited performance compared to US chips. Dylan Patel from SemiAnalysis noted that Nvidia's chip is actually better in practice due to superior memory than Huawei's 910B. Meanwhile, Huawei is struggling to produce the complex processors in sufficient quantity to meet demand. Most Chinese AI companies have built their models on Nvidia's ecosystem and software, and switching to Huawei's infrastructure would be time-consuming and costly. This highlights the complex dynamics in the global chip market and the challenges companies face in times of geopolitical tensions.