China’s strategy in AI competition with the US – large chip clusters, cheap energy

It is well known that Chinese semiconductors designed for artificial intelligence cannot compete with US companies. Nvidia. Still, China continues to develop advanced AI models, many of which are run on domestically produced chips.

China’s secret? China’s tech champion Huawei’s vast quantities of cheap energy and giant chip clusters are underpinning AI advances in competition with the US

“China views AI as a strategic technology for national security and economic security, so it is striving for self-sufficiency across the AI ​​stack,” Wendy Zhang, senior analyst at Mercator Research Institute for China (MERICS), told CNBC.

Questions have swirled over the country’s ability to compete in AI, with the world’s second-largest economy cut off from certain technologies by U.S. regulations and the Chinese government choosing to avoid using NVIDIA chips.

Despite these geopolitical challenges, domestic technology companies alibaba DeepSeek has successfully developed and released high-performance AI models, many of which are trained on its own chips.

When it comes to semiconductors needed to train and run AI models and applications, Nvidia’s graphics processing units (GPUs) are considered the gold standard. However, U.S. export restrictions have prevented Nvidia from shipping its cutting-edge chips to China.

Under an agreement with the White House this year, Nvidia received permission to market its H20 product, a downgraded chip designed for China. However, the Chinese government has reportedly encouraged Chinese companies to avoid Nvidia products and instead use chips designed for domestic companies.

Here comes Huawei, one of China’s most famous technology giants, developing the Ascend series of chips. But on a per-chip basis, Huawei doesn’t compete with Nvidia. Rather, Huawei’s advantage comes from its ability to link many of these chips together into high-performance “clusters” that can compete with Nvidia.

One of those products is the Huawei CloudMatrix 384, which connects 384 Ascend 910C chips to provide performance comparable to Nvidia’s GB200 NVL72, one of the most advanced systems. Nvidia’s system uses 72 GPUs, while Huawei’s product uses its own Ascend chip, which has five times as many.

“This strategy relies on high-speed, potentially optical interconnects to quickly move data between large clusters. This setup suits China’s current strengths because it does not require top-end chips,” Brady Wang, associate director at Counterpoint Research, told CNBC.

The downside to Huawei’s system is that using more chips means power consumption also increases significantly. This is where China’s energy superiority over the United States comes into play.

“Solutions like CloudMatrix are less power efficient than Nvidia systems, but here China benefits from abundant and cheap energy,” said MERICS’ Zhang.

“China is investing heavily in green energy such as solar and wind power. It is also rapidly expanding its nuclear infrastructure, so it can rely on cheap energy when building its AI infrastructure.”

The Chinese government and local governments are trying to support this effort. Many cities across China, from Shanghai to technology hub Shenzhen, are offering subsidies or “vouchers” to companies looking to rent computing power to reduce costs.

The Financial Times reported this week that some local governments in China are offering subsidies to reduce electricity costs for data centers using domestically produced chips.

“Accelerators with fewer advanced processes consume more power, but China compensates for this with diverse energy sources, renewables such as nuclear and solar power, and low rents and financing, allowing it to fund and operate large clusters despite chip-level inefficiencies,” said Counterpoint Research’s Wang.

The question is, as AI semiconductors advance, will Huawei and SMIC be able to catch up to Nvidia and TSMC, given Chinese companies’ limited access to key technologies?

“One of the key constraints on this strategy is whether China has the ability to produce enough chips domestically to make up the difference and catch up as NVIDIA and others continue to improve as well,” Hanna Dohmen, senior research analyst at Georgetown’s Center for Security and Emerging Technologies (CSET), told CNBC.

“China is working hard to increase its semiconductor manufacturing capacity and production capacity, but export restrictions imposed by the United States and its allies on semiconductor manufacturing equipment continue to cause significant delays.”