China tech sector leans on Nvidia chips as AI compute shortage persists

Traders Union
All News
Financial News
China relies on Nvidia for AI chips

Highlights

Chinese tech firms purchased over 1 million Nvidia chips in 2024 amid a domestic AI chip supply of 3.5 million and demand above 5 million.
Alibaba, ByteDance, and Tencent may more than double combined capital expenditure to about Rmb700 billion ($104 billion) in 2024 as AI model and infrastructure investment intensifies.
Tencent, ByteDance, and Alibaba are securing overseas Nvidia GPUs, including the B300, to offset domestic shortages, sustaining opportunities for U.S. chipmakers in China despite self-sufficiency targets.

Chip demand outpaces domestic AI supply

As reported by the Financial Times, Chinese technology companies have bought more than 1 million Nvidia gaming and workstation chips this year as domestic supply fails to keep up with surging AI demand.

Washington began restricting sales of advanced AI processors to China four years ago, and Beijing instructed technology companies two years ago to reduce reliance on Nvidia. Even so, Nvidia remains central to parts of China’s AI build-out because companies need additional computing capacity while local alternatives scale up.

Industry insiders estimate domestic supply could reach as many as 3.5 million AI chips this year, including about 1 million manufactured overseas, but total demand is expected to far exceed 5 million. China has more than doubled production of domestic AI chips this year, led by groups including Huawei, Alibaba, Baidu and Cambricon, yet spending on models, infrastructure and data centres is rising faster still.

To bridge the gap, companies are turning to Nvidia chips such as the RTX Pro 6000D, RTX Pro 5000 and RTX 5090, which comply with U.S. export controls and can be repurposed for some AI workloads. These processors trail leading Chinese AI chips such as Huawei’s Ascend 910C in most areas, but they are still capable of handling simpler tasks including the prefilling stage of inference.

Nvidia says China’s domestic makers of workstation chips and cards have seen unprecedented growth since 2022, while U.S. companies remain constrained by export controls and China’s own limits on U.S. imports. Another reason for continued demand is Nvidia’s Cuda software ecosystem, which many Chinese technology groups used before shifting toward domestic alternatives such as Huawei’s Cann, which is seen as less easy to use.

AI investment surge widens strategic pressure

The supply squeeze is intensifying as major Chinese technology companies sharply increase AI-related spending. Alibaba, ByteDance and Tencent are all pushing advanced models and agentic tools this year, lifting demand for tokens, the basic units of data processed and generated by AI models.

People with direct knowledge of the matter say combined capital expenditure from those three groups could more than double to about Rmb700 billion, or $104 billion, this year. Morgan Stanley analysts project China’s total spending on computing power this year at $193 billion, equivalent to about 21 per cent of U.S. spending.

Chinese companies are also looking abroad for capacity they cannot secure at home. Tencent has signed with Oracle to lease 100,000 GPUs across data centres in south-east Asia, while ByteDance and Alibaba are pursuing similar large-scale, long-term leasing agreements, according to people familiar with their plans.

Many of those overseas facilities are equipped with Nvidia’s latest chips, including the B300, which can help offset the shortage of advanced onshore computing for AI training. The widening gap between demand and local supply is complicating Beijing’s push to build a self-sufficient AI industry, while also preserving an opening for U.S. chipmakers led by Nvidia in the Chinese market.

Our earlier article on Meta’s Muse Wonder and Muse Wonder Full launch explained how the company is moving into frontier models, pairing higher benchmark targets with premium pricing via its API and subscription tiers. We also noted Meta’s focus on “active memory” features for longer-running assistant use cases and the growing capital spending required to train and deploy these systems at scale.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.