China orders $35B of Nvidia chips to bypass US export bans

US export bans didn't stop China's AI push. Tech giants just ordered $35 billion of Nvidia's backup chips, and supply is already sold out.

Business · Source: Reuters

What happened

ByteDance, Alibaba, and Tencent just placed a massive $35 billion order for Nvidia chips. They want more than 3 million RTX PRO 5500 graphics cards. These are workstation chips. They are designed for powerful desktop computers, not massive data center clusters. But they are the most powerful Nvidia processors still legal to sell in China.

The US government banned Nvidia from selling its top-tier B300 chips to Chinese buyers. Exports of the Hopper-based H200 are also heavily restricted and reviewed on a case-by-case basis. The goal was to starve China of cutting-edge AI computing power. Instead, Chinese tech giants are buying up every export-compliant chip they can find. They plan to use these cards primarily for AI inference. This means running trained models to generate chatbot responses, which requires less raw power than training massive datasets.

Nvidia cannot even fill the order. The company reportedly told partners they might only receive 5 to 10 percent of what they requested. Nvidia indicated it can produce no more than 3 million RTX PRO 5500 cards globally in 2027. However, total orders from China alone have already exceeded 6 million. Demand is double the entire global supply.

Key facts

Why it matters

This changes the hardware landscape for anyone building AI products today. If you rely on mid-tier or workstation GPUs for inference, prepare for a massive supply shock. China is effectively buying up the entire global supply of these chips. Prices for secondary hardware will likely spike as availability drops to zero. Cloud providers who failed to secure hardware allocations early will pass these rising costs directly to you.

The second-order effect is a deeply fractured global AI ecosystem. US sanctions forced Chinese companies to build massive inference clusters out of weaker, secondary chips. This limitation drives intense innovation in distributed computing and software optimization. Chinese engineers have to squeeze every drop of performance out of inferior hardware. Washington's curbs also push Chinese firms toward domestic suppliers. Beijing is actively encouraging the use of homegrown chips. Huawei has already gained market share. While Chinese chipmakers struggle to match leading American products due to separate equipment bans, this guaranteed domestic demand buys them time and capital to catch up.

For builders

Prepare for severe workstation GPU shortages

China just bought the entire 2027 global supply of Nvidia's top workstation cards. If you build local AI rigs or rely on mid-tier cloud GPUs, expect long delays and steep price hikes. Smaller cloud providers who lack priority access will lose out.

Optimize your software for inference efficiency

Chinese tech giants are forced to run AI inference on less powerful hardware. This will drive massive open-source software improvements for running models on weaker chips. Builders who adopt these new optimization techniques will save significant money on compute costs.

Watch the rise of alternative hardware

The complete sellout of Nvidia hardware forces buyers to look elsewhere. Domestic Chinese suppliers like Huawei are gaining market share rapidly. Builders should monitor these alternative ecosystems as they mature and potentially offer cheaper compute options in the future.

My take

I look at this and laugh. America tried to kill China's AI industry by banning the best chips. All it did was turn Nvidia's backup hardware into a $35 billion gold rush. You cannot regulate demand out of existence. You only force it to find another way. Now Nvidia is sold out of chips it hasn't even built yet, and China is paying full price for scraps.

Original reporting: Reuters. This is my rewrite and opinion.

More AI news for builders