Global GPU Computing News【20250512】

1. China’s GPU Self-Sufficiency Rate Expected to Reach 82% in Three Years
   Morgan Stanley reports that China’s AI GPU self-sufficiency rate will surge from 34% in 2024 to 82% by 2027, driven by accelerated domestic substitution. The STAR Market AI ETF (588930) saw net inflows for four consecutive days, with daily turnover exceeding ¥50 million, reflecting strong market confidence in AI computing.

2. Beijing Subsidizes Domestic GPU Purchases
   Beijing released a computing infrastructure plan offering financial incentives for companies procuring homegrown GPU chips. The plan aims to build intelligent computing clusters and achieve 45 EFLOPS of AI computing capacity by 2025.

3. Moore Threads Deploys 1,000-Card Clusters for Major AI Models
   Moore Threads’ KUAE data center solution has completed large-scale cluster deployments supporting Llama, GPT, and other mainstream models. It achieved a 91% linear acceleration ratio and passed CAICT certification, lowering barriers for domestic AI adoption.

4. SMIC Q1 Net Profit Soars 166.5%
   SMIC reported a 30% YoY revenue growth and ¥1.356 billion net profit in Q1, fueled by semiconductor demand, boosting confidence in domestic chip manufacturing.

5. Tencent Cloud Launches AI Suite, Cuts Training Time
   Tencent Cloud introduced an integrated software-hardware solution (StarNet 2.0, TurboFS) to address GPU cluster inefficiencies, reducing trillion-parameter model training from 50 days to 4 days.

6. Imagination Unveils E-Series GPU with 400% Edge AI Boost
   The new E-series GPU features a neural core (200 TOPS INT8) and burst processing, improving power efficiency by 35% for edge applications like autonomous driving. It launches in fall 2025.

7. **NVIDIA CEO: China’s AI Market to Hit $\mathrm{<span\; style="color:\#FFFFFF;">50</span>}\mathrm{<span\; style="color:\#FFFFFF;">B</span>}<span\; style="color:\#FFFFFF;">\ast </span><span\; style="color:\#FFFFFF;">\ast </span>\mathrm{<span\; style="color:\#FFFFFF;">J</span>}\mathrm{<span\; style="color:\#FFFFFF;">e</span>}\mathrm{<span\; style="color:\#FFFFFF;">n</span>}\mathrm{<span\; style="color:\#FFFFFF;">s</span>}\mathrm{<span\; style="color:\#FFFFFF;">e</span>}\mathrm{<span\; style="color:\#FFFFFF;">n</span>}\mathrm{<span\; style="color:\#FFFFFF;">H</span>}\mathrm{<span\; style="color:\#FFFFFF;">u</span>}\mathrm{<span\; style="color:\#FFFFFF;">a</span>}\mathrm{<span\; style="color:\#FFFFFF;">n</span>}\mathrm{<span\; style="color:\#FFFFFF;">g</span>}\mathrm{<span\; style="color:\#FFFFFF;">p</span>}\mathrm{<span\; style="color:\#FFFFFF;">r</span>}\mathrm{<span\; style="color:\#FFFFFF;">e</span>}\mathrm{<span\; style="color:\#FFFFFF;">d</span>}\mathrm{<span\; style="color:\#FFFFFF;">i</span>}\mathrm{<span\; style="color:\#FFFFFF;">c</span>}\mathrm{<span\; style="color:\#FFFFFF;">t</span>}\mathrm{<span\; style="color:\#FFFFFF;">s</span>}\mathrm{<span\; style="color:\#FFFFFF;">C</span>}\mathrm{<span\; style="color:\#FFFFFF;">h</span>}\mathrm{<span\; style="color:\#FFFFFF;">i</span>}\mathrm{<span\; style="color:\#FFFFFF;">n</span>}\mathrm{<span\; style="color:\#FFFFFF;">a</span>}\text{<span style="color:\#FFFFFF;">’</span>}\mathrm{<span\; style="color:\#FFFFFF;">s</span>}\mathrm{<span\; style="color:\#FFFFFF;">A</span>}\mathrm{<span\; style="color:\#FFFFFF;">I</span>}\mathrm{<span\; style="color:\#FFFFFF;">m</span>}\mathrm{<span\; style="color:\#FFFFFF;">a</span>}\mathrm{<span\; style="color:\#FFFFFF;">r</span>}\mathrm{<span\; style="color:\#FFFFFF;">k</span>}\mathrm{<span\; style="color:\#FFFFFF;">e</span>}\mathrm{<span\; style="color:\#FFFFFF;">t</span>}\mathrm{<span\; style="color:\#FFFFFF;">w</span>}\mathrm{<span\; style="color:\#FFFFFF;">i</span>}\mathrm{<span\; style="color:\#FFFFFF;">l</span>}\mathrm{<span\; style="color:\#FFFFFF;">l</span>}\mathrm{<span\; style="color:\#FFFFFF;">r</span>}\mathrm{<span\; style="color:\#FFFFFF;">e</span>}\mathrm{<span\; style="color:\#FFFFFF;">a</span>}\mathrm{<span\; style="color:\#FFFFFF;">c</span>}\mathrm{<span\; style="color:\#FFFFFF;">h</span>}$50 billion in three years, calling it a “huge loss” to ignore. Morgan Stanley estimates China will claim 20% ($48B) of global cloud AI by 2027.

8. Meta, OpenAI Drive Demand for 10,000-Card Clusters
   Training Llama 3 required 24,000 H100 GPUs, while Sora used over 10,000 GPUs. Surging model complexity is pushing global clusters toward 10,000+ GPU scale, with interconnect tech as a bottleneck.

9. Compute Networks Prioritize Storage-Network Synergy
   China’s National Data Bureau plans a unified computing network to balance regional resources. JD Cloud’s vGPU platform improves utilization and cuts costs.

10. AI Capex Recovery Spurs Opportunities
    Microsoft and Meta’s earnings signal renewed AI spending, with policy tailwinds boosting sectors like AI agents and humanoid robots, per East China Securities.

Key Takeaways: This week highlighted China’s rapid GPU substitution (policy, tech, capital) and global competition in compute power. Innovations like Moore Threads’ clusters and Imagination’s edge GPUs dominate, while challenges persist in cluster efficiency, cost control, and infrastructure coordination.