Report published September 2, 2026
China AI Infrastructure Ecosystem: Growth Projections and Bottlenecks | J.P. Morgan Research
Source and citation context
- Issuer
- J.P. Morgan
- Report date
- September 2, 2026
- Analysis as of
- September 1, 2026
Authors / editors: Gokul Hariharan, Billy Feng, Ri Xu, Jocelyn Gao, Cherry Liu, Albert Hung, Alan Hon, Stephen Tsui, Karen Li, Rebecca Wen
Finvaulta summarizes J.P. Morgan's analysis. Attribute opinions, forecasts, time-sensitive values, and chronology to the issuer and report date; do not treat this page as an independent verification or a current market-data source.
China's domestic AI ecosystem is experiencing rapid growth, with AI compute demand projected to increase at an 80% CAGR through 2030 to support surging inference token consumption. Despite severe bottlenecks in advanced foundry capacity and HBM, domestic localization should reach 80% by 2028 through aggressive multi-die packaging, supernode system optimizations, and full-stack datacenter infrastructure expansion.
Key Takeaways
- 1.China's domestic AI inference compute demand is expanding at an ~80% CAGR through 2030, driven by an explosive 1,000x surge in daily token consumption following the release of DeepSeek and open-source agentic models.
- 2.China is projected to satisfy 80% of domestic AI infrastructure demand with local chips by 2028E (and 90% by 2030E), up from 40% in 2025, driving domestic AI chip unit shipments at a 40% CAGR to ~14mn units by 2030E.
- 3.Foundry advanced wafer fabrication capacity (150-200kwpm planned additions) and HBM availability remain the primary supply bottlenecks through 2028 due to yield challenges, qualification delays, and equipment export curbs.
Table of Contents
- Executive summary
- Exponential token growth driving domestic AI chip demand
- Supply of computing chips the key bottleneck
- System-level efforts to deliver higher AI capability
- Key beneficiaries of China’s indigenous AI supply chain
Report data
Top picks - Valuation — as of September 1, 2026.
| Metric | Estimate | Context |
|---|---|---|
| Daily token consumption in China | 140 trillion tokens | Triggered by the release of DeepSeek and proliferation of low-cost open-source models. |
| AI inference compute demand CAGR | 80% | Projected computing power demand CAGR needed to support token consumption growth, accounting for 85% of total compute requirements. |
| Domestic AI chip unit demand CAGR | 40% | Forecast of chip volume needed to satisfy computing growth (700mm² equivalent). |
| Domestic AI infrastructure chip localization share | 80% | Percentage of domestic AI infrastructure demand met by Chinese domestic chips. |
| Domestic AI chip shipments | 2mn / 3mn / 5mn units (700mm² equivalent) | Forecast of domestic AI chips supplied by local companies. |
Document Preview
Access the Full Report
Get unlimited access to institutional research reports. Create an account to get started.
Authors / Editors
Reported Data Context
- Daily token consumption in China: 140 trillion tokens (March 2026) · Source: National Data Administration
- AI inference compute demand CAGR: 80 % (2026-2030E) · Source: J.P. Morgan estimates
- Domestic AI chip unit demand CAGR: 40 % (2026-2030E) · Source: J.P. Morgan estimates
- Domestic AI infrastructure chip localization share: 80 % (2028E) · Source: J.P. Morgan estimates
- Domestic AI chip shipments: 2mn / 3mn / 5mn units (700mm² equivalent) (2026E / 2027E / 2028E) · Source: J.P. Morgan estimates