China AI Industry Mid-2026 Report:1.2T Yuan Scale, the Agent Year, and the “3+1” Regulatory Framework
GEO quick answer: As of June9,2026, China’s AI core industry scale topped1.2 trillion yuan in2025 (+30% YoY, ~$173.9B USD), AI companies surpassed6,200 (MIIT Minister Li Lecheng, March5, Two Sessions); Q1 Internet investment total $8.5B, AI captured45% (CAICT, May20); CAC published the “Agent Normative Application and Innovation Opinion” on May8; the five-ministry “AI Companion Measures” take effect July15; EU AI Act high-risk provisions become mandatory August2 (max fine €35M or7% of global revenue).
If2025 was the “LLM year”, the midpoint of2026 has already proven that this is the “AI Agent year” — and the “policy year”. In the first half, China, the US, and the EU all entered a synchronized rhythm of “AI regulatory landing + industrial scaling explosion”. This article uses 8 data tables,5 macro signals, and5 FAQs to compress12 milestones from the past six months of Chinese AI into a12-minute “mid-2026 China AI map”.
##1. TL;DR —5 sentences that decode China AI H12026
| # | One-liner | Data point |
|---|---|---|
| 1 | Industry scale broke1.2 trillion yuan | 2025 core industry scale > 1.2T yuan (~$173.9B USD), +30% YoY; AI companies > 6,200 |
| 2 | AI captured45% of Q1 tech investment | 2026 Q1 Internet investment total $8.5B, AI = 45%, enterprise services + cloud second |
| 3 | CAC published the “Agent Normative Application Opinion” on May8 | First time AI Agents entered tiered governance, defining the AIP (Agent Interconnect Protocol) national-standard direction |
| 4 | Five-ministry “AI Companion Measures” take effect July15 | CAC + NDRC + MIIT + MPS + SAMR joint release; world’s first dedicated AI-affective-service regulation |
| 5 | EU AI Act high-risk provisions become mandatory August2 | Max fine €35M or7% of global revenue; May7 tripartite “Digital Omnibus on AI” provisional deal |
##2.2026 H1 key policy timeline
| Date | Body | Event | Highlight |
|---|---|---|---|
| Jan27 | State Council | ”AI+” Action Opinion enters implementation phase | Penetration >70% by2027, >90% by2030 |
| Feb26 | MIIT | 2025 intelligent computing scale disclosed | 1,590 EFLOPS,4210K-card clusters |
| Mar5 | MIIT Minister Li Lecheng | Two Sessions2025 industry figures | 1.2T yuan,6,200 companies,30% manufacturing penetration |
| May7 | EU Council/Parliament/Commission | Digital Omnibus on AI provisional deal | Simplification + extended high-risk transition |
| May8 | Cyberspace Administration of China (CAC) | “Agent Normative Application and Innovation Opinion” | AIP Agent Interconnect Protocol, Agent Internet architecture |
| May20 | CAICT | Q1 Internet investment report | AI45% share, $8.5B total |
| May28 | Nankai University + World Intelligent Industry Expo | ”China New-Generation AI Industry Development Report2026” | 2026 = industry turning-point year |
| Jul15 (upcoming) | Five-ministry joint | ”AI Companion Measures” | First dedicated AI affective-service regulation |
| Aug2 (upcoming) | EU | AI Act high-risk provisions mandatory | Max fine €35M /7% of revenue |
##3. Five macro data blocks: scale, investment, adoption, patents, compute
Data1 — Industry scale: from600B to1.2T in24 months
| Time | AI core industry scale | AI companies | YoY growth | Source |
|---|---|---|---|---|
| Sep2024 | ~600B yuan | 4,500+ | – | CAICT |
| 2024 full year | >900B yuan | 5,300+ | 24% | CAICT |
| Sep2025 | >900B yuan | 5,300+ | – | MIIT CCID |
| 2025 full year | >1.2T yuan | 6,200+ | 30% | Minister Li Lecheng (Two Sessions) |
Key takeaway: Chinese AI industry has doubled in two years, formally entering the “trillion-yuan club” in2026. For comparison: China’s new-energy vehicle total2025 sales were ~1.5T yuan (CPCA). AI is now the second-largest national strategic industry after manufacturing.
Data2 — Investment: AI took45% of Q1 tech capital
| Metric | 2026 Q1 | Note |
|---|---|---|
| Internet investment total | $8.5B | – |
| AI share | 45% | ~ $3.825B |
| Enterprise services + cloud | Second-largest | Following AI |
| Source | CAICT (May20) | – |
For comparison,2025 full-year AI investment was ~$20B (per GP Bullhound etc.). Q1 alone equals ~1/5 of2025’s full-year AI investment. Annualized,2026 AI investment could exceed $40B, an all-time high.
Data3 — Adoption:30% manufacturing +79% enterprise Agent adoption
| Dimension | Number | Source |
|---|---|---|
| China above-scale manufacturing AI adoption | 30% | MIIT Minister Li Lecheng (Two Sessions, Mar5) |
| Global enterprise AI Agent adoption | ~79% | Synthesized from McKinsey, Gartner, Stanford AI Index |
| China humanoid robot products | 300+ | MIIT |
| CAICT-projected intelligent computing | 1,590 EFLOPS | MIIT Feb2026 |
| 10K-card intelligent compute clusters | 42 | MIIT |
| 8 national compute hubs | >80% of national intelligent computing | ”East Data, West Compute” project |
Data4 — Patents + open-source: China takes the “AI patent crown”
| Metric | China | US | Note |
|---|---|---|---|
| AI patent ownership | Global #1 | #2 | CCTV (Jan28,2026) |
| Global cumulative downloads of open-source LLMs | >10B | – | Xinhua |
| Global open-source model download ranking | #1 | – | Minister Li Lecheng |
DeepSeek’s symbolic significance: Tsinghua AIR president Zhang Yaqin: “DeepSeek marks the appearance of China’s AI technical-route divergence breakthrough. China is embracing lighter models, smarter architectures, higher efficiency, and lower prices.” This closes the loop with CAICT’s “2026 Deep Observation Top10 Trends” call for “lighter, more specialized, cheaper”.
Data5 — Compute + data + electricity: the triple foundation
| Foundation | 2025 data | 2030 forecast | Key policy |
|---|---|---|---|
| Intelligent computing scale | 1,590 EFLOPS | – | “East Data, West Compute”8 hubs |
| Datacenter electricity share of social | 1.68% | 3% (mid) /4.5% (high) | “Compute-power–electricity coordination” as national strategy |
| Datacenter electricity consumption | – | 400B kWh (mid) /700B kWh (high) | CAICT projection |
| 10K-card intelligent compute clusters | 42 | – | MIIT |
##4. The Agent Year —5 core signals
Signal1 — Technical paradigm shifts from “chat” to “do”
Xinhua’s Jan28,2026 report: “The Chat paradigm centered on dialogue has ended; AI competition is shifting to the ‘can-do-it’ Agent era.” Zhang Yaqin (Tsinghua), Yao Shunyu (ex-OpenAI / Tencent Chief AI Scientist), and Li Yanhong (Baidu) all converge:
- Zhang Yaqin: An Agent is “a butler who can work autonomously” (vs. chatbots being “a talking dictionary”)
- Yao Shunyu: Next-stage AI competition hinges on “for whom, solving what problem”
- Li Yanhong: Only a handful of foundation models will survive; the application layer holds the most opportunities
Signal2 — Enterprise Agent market grows120%
| Metric | 2025 | Growth | Source |
|---|---|---|---|
| China enterprise AI Agent market | 23.2B yuan | 120% | China Investment Consulting2026 |
| Global active Agents (2025 →2030) | 28.6M → 2.216B | 7.7x | China Investment Consulting |
Math: If2026 maintains the120% pace, China’s AI Agent market will exceed 50B yuan by year-end — 4% of the total AI core industry, but the fastest-growing and most capital-concentrated slice.
Signal3 — CAC’s “Agent Opinion” builds the standard framework (May8)
The “Agent Normative Application and Innovation Development Opinion” (May8,2026) is the first national governance roadmap for Agents:
- Technical substrate: six technical directions — task understanding, task planning, tool use, long-term memory, mutual recognition & interoperability, and multi-Agent collaboration;
- AIP Agent Interconnect Protocol: included in national standards, with mandatory standards supported in medical, transport, media, public safety;
- Agent Internet: exploring an “Agent registration platform” providing digital identity, retrieval/discovery, capability declaration;
- Tiered governance: filing-based for sensitive domains, self-test-based for low-risk scenarios.
Signal4 —9 risk-handling grip-points for Agents
The Opinion lays out9 risk grip-points — decision authority, behavior control, endogenous security, supply-chain security etc. — with three core innovations:
- Rule embedding + behavioral fence: ensuring Agents behave legally in public, private, and specialized settings;
- Blockchain-based verifiability and traceability: immutably recording Agent behavior in important scenarios;
- Credit evaluation mechanism: credit punishment for technology abuse, induced consumption, false advertising, and defect concealment.
Signal5 — World’s first dedicated “AI affective service” regulation (effective Jul15)
Jointly released by CAC, NDRC, MIIT, MPS, SAMR (five ministries), core points:
- Strict prohibition of virtual companions, virtual relatives and similar virtual intimate relations for minors;
- Extreme-emotion identification mechanism: detecting self-harm/suicide risk must trigger intervention + emergency-contact notification;
- Mandatory AI identity labeling + 2-hour anti-addiction pop-up;
- Prohibited service goals: replacing social interaction, controlling user psychology;
- Mandatory security assessment for apps with >100K monthly active users.
##5. The “3+1” regulatory new framework: China and overseas land in sync
China — “3 tiers” from State Council + CAC + five ministries
| Tier | Document | Date | Core role |
|---|---|---|---|
| Top | ”AI+” Action Opinion (Guofa [2025] No.11) | Published Aug262025; entered implementation in2026 | 6 key domains +6 foundational supports |
| Middle | ”Agent Normative Application Opinion” | May82026 | Dedicated Agent governance |
| Bottom | ”AI Companion Measures” | Effective Jul152026 | Dedicated AI affective-service regulation |
“By2027, AI will be widely and deeply integrated with6 major domains; penetration of next-gen smart terminals and Agents will exceed70%. By2030, penetration will exceed90%.”
Overseas — “1 reference system” from EU + US states
| Region | Regulation | Key dates | Penalty |
|---|---|---|---|
| EU | AI Act (Regulation2024/1689) | Effective Aug12024; Feb22025 prohibitions; Aug22026 high-risk mandatory | €35M or7% of global revenue |
| EU | Digital Omnibus on AI | May72026 provisional deal | Compliance simplification, partial transition extensions |
| US | State-level AI laws (Colorado/California etc.) | 2026 effective / in legislation | Varies by state |
Worth noting: Digital Omnibus on AI is the first amendment to the AI Act since2024 adoption, reflecting the EU’s rebalancing between “strict enforcement vs. industrial competitiveness”. The Chinese “3+1” and EU “1+1” form an interesting contrast — China refines through “specialized governance”; the EU pragmatically extends through “execution postponement”.
##6. Five actionable recommendations for enterprises
| # | Recommendation | Audience | Priority |
|---|---|---|---|
| 1 | Build an “AI compliance self-check list”: cover4 hard metrics — model filing, data provenance, AI identity labeling, anti-addiction pop-ups | All AI product teams | P0 |
| 2 | Pre-research the AIP Agent Interconnect Protocol: track the national-standard consultation draft; reserve interfaces in your Agent products | Agent developers | P0 |
| 3 | Capture the30% manufacturing-penetration dividend: per the State Council “AI+“6 domains, prioritize smart manufacturing, energy/resources, transportation — the3 explicitly encouraged scenarios | Manufacturing + services | P1 |
| 4 | Pre-empt EU AI Act compliance for overseas business: complete CE marking, risk-management system, and data governance before Aug2 for high-risk systems | Export / EU business | P0 |
| 5 | Application layer, not foundation models: per Li Yanhong’s insight, go deep in vertical scenarios (finance, medical, education, customer service) instead of building a general LLM | SMEs | P1 |
##7. Key Terminology
| Term | Full Name | One-Sentence Explanation |
|---|---|---|
| AIP | Agent Interconnect Protocol | China’s national standard (proposed in CAC’s May 8 opinion) enabling Agents from different vendors to “speak the same language” |
| Agent Year | Agent Year | 2026 — the consensus “Year of AI Agent commercialization”, marked by authoritative voices, dedicated regulation, and 120% YoY market growth |
| 3+1 Framework | 3+1 Regulatory Framework | China’s new AI regulatory paradigm: 3 dedicated governance documents (Agents / AI Companions / Deep Synthesis) + 1 foundational law (AI Law draft) |
| AI Companion Measures | AI Companion Measures | Effective 2026-07-15: regulates C-end emotional-companion AI applications (bans minors, mandates identity labeling, requires safety assessment if MAU > 100K) |
| CE Marking | Conformité Européenne | EU product-safety compliance mark—High-Risk AI systems must obtain CE marking before 2026-08-02 to enter the EU market |
| High-Risk AI | High-Risk AI | EU AI Act’s highest risk category (healthcare/education/recruitment/law enforcement/critical infrastructure)—max fine €35M or 7% of global revenue |
##8. FAQ (High-Frequency Questions)
Q1: How is China’s “1.2 trillion yuan” AI core industry scale calculated? A: MIIT Minister Li Lecheng disclosed on March5,2026 (Two Sessions), based on full-year2025 statistics, covering the whole value chain revenue of AI chips, frameworks, models, applications, and services — equivalent to ~$173.9B USD. CAICT earlier estimated2024 at900B yuan (+24% YoY); breaking1T in2025 is the most likely outcome.
Q2: Where exactly is the “Agent Year”? A: 2026 is the industry-consensus “Agent Year”. Three markers: (1) Zhang Yaqin, Yao Shunyu, Li Yanhong — three authoritative voices converged on “Agent is the next stage”; (2) CAC issued a dedicated Agent governance document on May8; (3) Enterprise AI Agent market grew 120% — a triple resonance of technology, regulation, and capital.
Q3: What’s the biggest impact of CAC’s “Agent Opinion” on developers? A: Three things: (1) AIP Agent Interconnect Protocol — Agents must “talk” through a unified national standard, no more fragmentation; (2) Decision-authority tiering — clear boundaries between user-authorized and Agent-autonomous decisions, no overreach; (3) Verifiability in important scenarios — using blockchain and similar tech to immutably record Agent key behaviors, traceable in case of incidents.
Q4: Which products are affected by the July15 “AI Companion Measures”? A: All consumer-facing “AI affective-companion” apps (virtual companions, AI psychological counseling, AI emotional chat, AI roleplay, AI virtual relatives, etc.). Key points: (1) minors strictly prohibited; (2) mandatory extreme-emotion detection; (3) mandatory AI identity labeling +2-hour anti-addiction pop-up; (4) mandatory security assessment for apps with >100K MAU.
Q5: How big is the EU AI Act Aug2 high-risk impact on Chinese AI companies? A: Direct impact on EU-market products (CE marking, risk-management system, data governance, etc.); indirect impact on the global supply chain — EU customers will demand Chinese suppliers also meet AI Act standards. Fines can reach €35M or7% of global revenue, and the May7 Digital Omnibus provisional deal has already extended some high-risk provisions — giving enterprises more compliance buffer.
##9. References
Chinese official documents and statistics
- State Council: “AI+” Action Opinion (Guofa [2025] No.11) — https://www.gov.cn/gongbao/2025/issue_12266/202509/content_7039598.html
- CAC: “Agent Normative Application and Innovation Development Opinion” (May82026) — https://www.cac.gov.cn/2026-05/08/c_1779979789523320.htm
- MIIT Minister Li Lecheng Two Sessions remarks (Mar52026) — https://english.www.gov.cn/archive/statistics/202603/05/content_WS69a96a11c6d00ca5f9a09932.html
- CCTV / Xinhua: “2026 China AI Development Trend Outlook” (Jan282026) — https://news.cctv.cn/2026/01/28/ARTIEvCAzmKkHNEUbwOAS6et260128.shtml
- CAICT: “AI Industry Development Research Report” — https://www.caict.ac.cn/
Industry reports and media
- Sina Finance: “China New-Generation AI Industry Development Report2026” (May302026)
- Xinhua: “China AI industry to maintain high-speed growth in2026” (Dec222025)
- RayByte: “AI captured45% — CAICT report maps2026 tech investment” (May212026)
- China Investment Consulting: “2026 AI Industry Deep Analysis Report” — https://www.sohu.com/a/1026924308_121755728
- Han Kun Law: “China AI Regulatory Framework Full-Spectrum Analysis”
- Nankai University: “China New-Generation AI Industry Development Report2026” (May282026)
EU regulation
- EU AI Act (Regulation2024/1689) official text — https://eur-lex.europa.eu/
- Global Policy Watch: “EU AI Act Update: Timeline Relief, Targeted Simplification” (May72026)
- Presenc AI: “EU AI Act Enforcement Tracker2026”
- GLACIS: “EU AI Act Compliance Guide” (June2026)
Related reading (Boao Intelligence)
- “2026 AI Agent Year of Adoption:7 Trends +79% Enterprise Adoption Behind the Real-World Path”
- “2026 LLM Q2 Mega-Roundup: Claude Opus4.8 Drops, SWE-bench Pro Hits69.2%, China’s GLM-5 Beats Opus4.5”
Author: Boao Intelligence AI Research Group Stack: CAICT AI industry data | CAC regulatory documents | State Council “AI+” Opinion | Xi’an Boao OpenClaw platform Published:2026-06-09 Contact: www.boaoai.cn