Navigating AI Inference Challenges: How China's Hyperscalers Are Pivoting Amid GPU Shortages

In 2025, China's AI infrastructure sector is undergoing a seismic shift as hyperscalers like Alibaba, Tencent, and Huawei pivot to CPU-optimized solutions and cross-border cloud partnerships to mitigate GPU shortages. U.S. export controls, supply chain bottlenecks, and the surging demand for AI workloads—particularly in large language models (LLMs)—have forced Chinese tech giants to rethink their strategies. This transition is not merely a stopgap measure but a strategic recalibration with significant implications for investors.

The GPU Shortage: A Catalyst for Innovation

The scarcity of advanced GPUs, such as NVIDIA's H20, has created a critical bottleneck for AI training and inference. H3C, a major server manufacturer and NVIDIA OEM partner, recently disclosed that its H20 inventory is nearly depleted, with shipments delayed until mid-2025. This scarcity has driven up secondary market prices for GPUs and intensified competition for limited supplies. Meanwhile, U.S. policies like the AI Diffusion Framework further restrict access to high-end chips for non-aligned countries, pushing Chinese firms to explore alternatives.

The financial impact is evident: Canalys reports that cloud infrastructure spending in Mainland China hit $11.6 billion in Q1 2025, a 16% year-on-year increase. AI workloads now account for triple-digit growth in cloud services, driven by the computational demands of foundation models. This surge has accelerated investments in hybrid architectures that combine domestic CPUs with limited GPU access and third-party accelerators.

CPU Optimization: A Strategic Shift

Chinese hyperscalers are increasingly adopting CPU-optimized solutions from domestic providers like Feiteng, Kunpeng, and Hygon. These chips, while not as powerful as NVIDIA's offerings, are viable for inference tasks and edge computing. Huawei's Kunpeng, for instance, has been deployed in mid-tier AI applications such as smart manufacturing and logistics, where its x86 architecture and RISC-V compatibility offer cost advantages.

The government's “Made in China 2025” policy, mandating 70% domestic chip usage in public-sector data centers, has further accelerated adoption. Feiteng and Hygon have shipped 10 million units combined, with TencentOS Server V3 now supporting their architectures. This policy-driven demand ensures a stable market for domestic CPUs, even as performance gaps with foreign GPUs narrow.

Liquid cooling is another enabler of this shift. Used in 22% of Chinese data centers, this technology allows for 80 kW rack densities, quadrupling the efficiency of air-cooled systems. It also supports faster processor refresh cycles, a critical factor in the rapidly evolving AI landscape.

Cross-Border Cloud Partnerships: Bridging the Gap

To circumvent GPU constraints, Chinese hyperscalers are forming cross-border partnerships with international cloud providers. Alibaba Cloud's expansion into Southeast Asia—opening a Global AI Capability Center (AIGCC) in Singapore and a third data center in Malaysia—exemplifies this trend. These hubs leverage more flexible access to U.S. technology under the AI Diffusion Framework, enabling Chinese firms to outsource compute-intensive tasks.

Collaborations with global open-source communities are also vital. Huawei's participation in the PyTorch Foundation and Alibaba's open-sourcing of its Qwen3 hybrid inference model highlight efforts to build interoperable ecosystems. These partnerships not only mitigate hardware limitations but also position Chinese firms to compete in global AI markets.

Financial Performance and Investment Opportunities

The AI infrastructure market in China is projected to grow at a 22.3% CAGR, driven by demand for AI accelerators and ASICs. Startups like Biren Technology and Moore Threads are raising capital to develop alternatives to NVIDIA, while established players like Alibaba Cloud maintain a 33% market share, with AI workloads driving 15% year-on-year growth.

Investors should focus on three key areas:
1. Domestic CPU Providers: Companies like Huawei (HWTIF) and Feiteng, benefiting from government mandates and R&D investments.
2. Cloud Infrastructure Leaders: Alibaba Cloud (BABA) and Tencent (TCEHY), which are expanding global AI centers and hybrid model capabilities.
3. Semiconductor Enablers: SMIC (SMIHF) and Xingsen Technologies, supplying advanced packaging and manufacturing for AI chips.

Conclusion: A Resilient Ecosystem

China's hyperscalers are demonstrating remarkable adaptability by leveraging CPU optimization, cross-border partnerships, and government support. While GPU shortages persist, these strategies are creating a resilient AI infrastructure ecosystem. For investors, the focus should be on firms that combine domestic innovation with global collaboration—those poised to lead in both self-reliance and international expansion. The “East Data, West Compute” initiative and the rise of open-source LLMs like DeepSeek's R1 signal a future where China's AI sector is not just surviving but thriving, offering compelling opportunities for those who recognize the strategic pivot.