China’s AI Strategy: Big Chip Clusters and Cheap Energy in the US Race

While it’s broadly acknowledged that Chinese-designed semiconductors for AI currently lag behind those of U.S. giant Nvidia, China has nonetheless made substantial strides in developing advanced AI models, many of which are being trained and run on domestically produced chips. This progression invites examination into the strategic advantages and technological workarounds enabling China to maintain its AI momentum amidst technological constraints.

The strategy hinges on leveraging the nation’s access to affordable energy and the deployment of expansive chip clusters, largely spearheaded by Huawei, China’s tech behemoth. These factors collectively underpin China’s AI ambitions, fueling its competitive drive against the United States.

“China is resolutely pursuing self-sufficiency across the entire AI stack, viewing AI as a cornerstone technology crucial for both national and economic security,” states Wendy Chang, senior analyst at the Mercator Institute for China Studies (MERICS), speaking to CNBC. This strategic perspective underscores the depth of China’s commitment to overcoming technological hurdles.

With U.S. restrictions limiting access to certain technologies, and Beijing seemingly signaling a preference away from Nvidia’s cutting-edge chips, questions surrounding China’s long-term competitive capacity in AI have inevitably surfaced. The geopolitical landscape is shaping technological trajectories and investment strategies within China’s AI ecosystem.

Despite these complex geopolitical dynamics, domestic Chinese tech firms, ranging from established players like Alibaba to emerging innovators such as DeepSeek, have successfully developed and launched competitive AI models, frequently training them on indigenous silicon solutions. This demonstrates a notable resilience and adaptability within the Chinese AI sector.

Nvidia’s GPUs are widely considered the gold standard for powering AI model training and inference. However, U.S. export controls restrict the shipment of Nvidia’s most advanced chips to China, forcing Chinese companies to seek alternative solutions.

While Nvidia received authorization to sell its H20 chip, a modified product tailored for the Chinese market, reports suggest that Beijing has been encouraging domestic firms to prioritize locally designed chips, further incentivizing the growth of China’s domestic semiconductor industry. The long-term implications of this policy shift on the broader AI landscape remains an area of keen observation.

Huawei, a key player in China’s technology sector, has emerged with its Ascend series of chips. Although Huawei’s individual chips may not directly rival Nvidia’s top-tier offerings in raw performance, the company has pioneered a strategy of linking these chips into high-performance “clusters.” This approach allows China to effectively compete with Nvidia by aggregating computing power at scale.

A prime example of this strategy is the Huawei CloudMatrix 384, a system that connects 384 of its Ascend 910C chips to deliver performance levels that rival Nvidia’s GB200 NVL72, one of its most advanced systems. Notably, while Nvidia’s system uses approximately 72 GPUs, Huawei’s solution employs five times as many of its Ascend chips. This disparity highlights a fundamentally different architectural philosophy.

“This clustering strategy leverages high-speed interconnects, potentially optical, to ensure rapid data transfer within these massive clusters,” Brady Wang, associate director at Counterpoint Research, explained to CNBC. The emphasis on advanced interconnect technology is crucial for mitigating performance bottlenecks and maintaining overall system efficiency given the increased chip count.

The increased power consumption associated with the cluster-based architecture represents a notable trade-off. However, this is where China’s distinctive energy landscape provides a significant comparative advantage.

“While solutions like the CloudMatrix are inherently less power-efficient than Nvidia systems, China capitalizes on its abundant and relatively inexpensive energy resources,” notes MERICS’ Chang. This access to cost-effective power is enabling the buildout and operation of massive AI infrastructure.

“China has invested heavily in renewable energy sources, including large-scale solar and wind projects. It is also aggressively expanding its nuclear energy capacity. This diversified and growing energy portfolio provides a stable and cost-effective foundation for supporting the energy-intensive demands of advanced AI infrastructure.”

Government support, both at the national and local levels, further amplifies China’s competitiveness. Numerous cities, including Shanghai and Shenzhen, have implemented subsidy programs to reduce the cost of computing power for companies utilizing domestic chips. These initiatives provide a direct incentive for companies to adopt and scale their AI operations leveraging Chinese hardware.

According to a report in the Financial Times, certain local governments in China are even offering subsidies specifically targeted at reducing electricity costs for data centers that predominantly use domestically produced chips. This underscores the comprehensive approach being taken to foster a thriving domestic AI ecosystem.

“Although less advanced process technologies used in domestic accelerators may result in higher power consumption, China’s strategic combination of diverse energy sources – including nuclear and renewable options – coupled with lower land and financing costs, allows for the funding and operation of large-scale compute clusters despite inefficiencies at the individual chip level,” Wang of Counterpoint Research points out. This highlights a systemic approach to overcoming technological limitations through strategic resource management.

The fundamental question remains: As AI-focused semiconductor technology continues to evolve, can Huawei and SMIC sustain pace with Nvidia and TSMC, particularly given ongoing restrictions on access to critical technologies?

“One of the primary challenges inherent in this strategy lies in China’s ability to domestically produce a sufficient volume of chips to both close and keep pace with the evolving performance gap as NVIDIA and others continue to innovate,” observes Hanna Dohmen, senior research analyst at Georgetown’s Center for Security and Emerging Technologies (CSET), in conversation with CNBC. This emphasizes that scaling production capacity is just as critical as technological advancement.

“China is investing significant resources in expanding its domestic semiconductor fabrication capabilities and capacity; however, it still lags significantly behind the leading manufacturers due to restrictions imposed by U.S. and allied nations on the export of advanced semiconductor manufacturing equipment.” This restriction on access to critical equipment is a significant bottleneck that requires substantial innovative solutions to fully overcome. Investment into alternative fabrication methodologies and materials science will be critical for enabling long-term technological independence.