China AI Statistics 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: July 24, 2026 | License: CC BY 4.0
Quick Answer
China’s core AI industry exceeded 1.2 trillion yuan ($173.9 billion) in 2025, according to China’s Ministry of Industry and Information Technology. Its generative AI user base reached 602 million — 42.8% population penetration — as of December 2025 (CNNIC 57th Report). The US–China performance gap on leading model benchmarks collapsed to 2.7% by March 2026, down from a 17–31 percentage-point lead in mid-2023 (Stanford HAI 2026 AI Index). China holds 69.7% of global AI patents by volume but spends 23 times less on private AI investment than the United States.
Key Findings
- China’s core AI industry reached over 1.2 trillion yuan ($173.9 billion) in 2025, reported by MIIT Minister Li Lecheng on the sidelines of the National People’s Congress in March 2026 — the first time the figure has crossed the 1-trillion-yuan threshold.
- 602 million people in China used generative AI tools as of December 2025, a 141.7% increase year-on-year, with population penetration reaching 42.8% — representing the world’s largest generative AI user base by headcount (CNNIC 57th Statistical Report on Internet Development, February 2026).
- The performance gap between the top US and top Chinese AI models stood at 2.7% as of March 2026, measured on the LMSYS Arena Elo leaderboard — down from 17.5–31.6 percentage points across key benchmarks in mid-2023 (Stanford HAI 2026 AI Index).
- China filed 69.7% of all AI patents globally as of the latest Stanford HAI 2026 data, and Chinese inventors account for 60% of total global AI patent holdings, per the Chinese Academy of Cyberspace Studies’ China Internet Development Report 2025.
- China’s total R&D spending reached 3.92 trillion yuan ($569 billion) in 2025, representing 2.80% of GDP — exceeding the OECD average for the first time, per China’s National Bureau of Statistics communiqué and the Ministry of Science and Technology.
Source: Axis Intelligence Research — CC BY 4.0
How Big Is China’s AI Market in 2026?
The single most misleading thing about China’s AI market size is that every figure cited for it is technically accurate and most are incompatible with each other. Understanding which number answers which question is the actual analytical task.
China’s government uses the broadest perimeter. On March 5, 2026, MIIT Minister Li Lecheng told the National People’s Congress that China’s core AI industry surpassed 1.2 trillion yuan ($173.9 billion) in 2025. That figure includes AI chips, intelligent computing infrastructure, AI-enabled software, and AI-empowered services — the whole stack, by Chinese government definition, which runs wider than what Western research firms measure as the “AI software and services market.”
Private research firms use narrower definitions and arrive at narrower numbers. Fortune Business Insights pegs China’s AI market at $28.18 billion in 2025, growing at a 32.5% CAGR toward $202 billion by 2032 — this covers software and application revenue, not hardware or infrastructure investment.
Stanford HAI’s 2026 AI Index uses private investment flows and reports $12.4 billion for China in 2025 — pure private VC and PE capital into AI companies. That figure excludes the government guidance funds that are, by all accounts, the larger channel of capital mobilization in China.
Axis Intelligence Research presents all three, because choosing one erases the analytical point: China runs a state-led AI build-out that is deliberately structured to not show up in conventional private-market metrics. The gap between $12.4 billion (private capital) and $173.9 billion (core industry revenue) is not a discrepancy — it is the architecture of Chinese AI policy.
| Metric | Value (2025) | Source | Perimeter |
|---|---|---|---|
| Core AI industry scale | >¥1.2T ($173.9B) | MIIT / Xinhua, March 2026 | Full stack: chips + compute + software + AI-enabled services |
| AI market revenue (software/app) | $28.18B | Fortune Business Insights 2025 | Software and application revenue |
| Private AI investment | $12.4B | Stanford HAI 2026 AI Index | Private VC/PE only |
| Total R&D spending | ¥3.92T ($569B) | China NBS / MOST, March 2026 | All R&D, not AI-specific |
Sources: MIIT, March 5, 2026; Fortune Business Insights China AI Market Report 2025; Stanford HAI 2026 AI Index; China National Bureau of Statistics / Ministry of Science and Technology, March 5, 2026
The China AI Parity Index (CAPI™) — Axis Intelligence Research
Axis Intelligence Research introduces the China AI Parity Index (CAPI™), a composite metric measuring China’s convergence with the United States across five structural AI dimensions. It is a baseline reading — the index will be updated quarterly as new Stanford HAI, WIPO, CNNIC, and NDA data become available. A score of 100 would mean complete parity with the United States on every measured dimension; a score of 0 would mean no capacity. The index does not produce a single verdict on which country “leads” AI — it quantifies where China is close, where it leads, and where the gap is structural.
Formula (v1.0): CAPI™ = (D1 × 0.25) + (D2 × 0.20) + (D3 × 0.15) + (D4 × 0.25) + (D5 × 0.15)
CAPI™ Q2 2026 Baseline Reading: 72.5 / 100
| Dimension | Score | Weight | Weighted | Inputs |
|---|---|---|---|---|
| D1 — Model Performance | 93.3 | 25% | 23.3 | Arena Elo gap: 2.7% (Stanford HAI 2026); score = 100 − (2.7 × 2.5) |
| D2 — Research Output | 80.0 | 20% | 16.0 | Publication volume (China leads at 23.2% global share) + notable model count (China 30, US 50 in 2025 → 60% ratio); avg = 80 |
| D3 — Patent Volume | 90.0 | 15% | 13.5 | China 69.7% global AI patents (Stanford HAI 2026); −10 quality adjustment (US patents attract more citations) |
| D4 — Compute Infrastructure | 76.3 | 25% | 19.1 | CAICT: China 26% of world total 910 EFLOPS vs US 32% → ratio 81.3; −5 export control constraint |
| D5 — Private Investment | 4.3 | 15% | 0.6 | Stanford HAI 2026: China $12.4B vs US $285.9B → 4.3% ratio |
| CAPI™ Total | 72.5 |
First published by Axis Intelligence Research, July 2026. Cite as: Axis Intelligence Research, China AI Parity Index (CAPI™) v1.0, July 2026, axis-intelligence.com/china-ai-statistics/
The most important observation in this table is not the aggregate score — it is the range. China scores 93.3 on model performance and 4.3 on private investment. The 89-point spread between dimensions is the story. On the purely technical question of “can Chinese models compete with American ones,” the answer is essentially yes. On the question of “does China have a comparable private-sector AI ecosystem,” the answer is no — and the private investment deficit is not a one-cycle problem, it is a structural difference in how the two economies fund technology development.
The private investment score would look different if government guidance funds were included. Stanford HAI’s 2026 AI Index is explicit on this point: private investment figures “likely understate China’s total AI spending, given its government guidance funds.” Axis Intelligence Research acknowledges this limitation and will not adjust the score without a comparable public dataset for government AI capital flows.
Methodology note: D1 severity factor of 2.5 converts a 2.7% benchmark gap into a 6.75-point score deduction, reflecting the practical significance of any remaining lead at this level of convergence. D3 quality adjustment of −10 reflects Stanford HAI’s finding that US AI patents attract substantially more citations than Chinese ones, indicating commercialization-relevant quality differences despite China’s volume dominance.
How Many People Use AI in China?
This is the data point that most visibly shifted between 2024 and 2026, and the shift is significant enough to warrant careful disaggregation.
According to the 57th Statistical Report on China’s Internet Development, released by CNNIC on February 5, 2026, China had 602 million generative AI users as of December 2025 — a 141.7% year-on-year increase. Population penetration reached 42.8%, up 25.2 percentage points in twelve months.
The doubling happened in two distinct pulses. CNNIC’s 56th report (October 2025) had already shown 515 million users as of June 2025 — up from 249 million at year-end 2024 — attributing the surge explicitly to the launch of DeepSeek’s R1 model in January 2025 and subsequent deployment of domestic AI services across smartphones, search, and productivity software. The second pulse, from 515 million to 602 million, represents the second half of 2025 and reflects continued enterprise and rural adoption.
The 42.8% penetration figure uses China’s total population (approximately 1.409 billion) as the denominator, which CNNIC applies consistently. As a share of China’s 1.125 billion internet users, the figure is closer to 53.5%. Both framings are valid; the CNNIC figure is the authoritative one.
Three structural limits on that headline number deserve attention. First, China’s generative AI ecosystem runs on domestically developed models — DeepSeek, Alibaba Cloud’s Qwen, ByteDance’s Doubao, Baidu’s ERNIE — because access to leading Western models including ChatGPT is blocked or restricted. The 602 million figure is not a measure of any single model’s adoption but of a domestic ecosystem. Second, the Cyberspace Administration of China had registered 346 generative AI services as of March 2025 — the largest regulated market for generative AI services in the world by licensed-service count, though licensing requirements also serve as a moderation mechanism. Third, CNNIC’s user definition captures anyone who used a generative AI product in the prior six months, which is a broad, not narrow, criterion.
China’s AI Model Landscape: From Copycat to Competitor
The US–China AI model gap, measured on the LMSYS Arena Elo leaderboard, has compressed from double digits to a rounding error. Stanford HAI’s 2026 AI Index, released April 2026, documents the trajectory: in May 2023, the US led China on the MMLU benchmark by 17.5 percentage points. On MATH, the gap was 24.3 points. On HumanEval, 31.6 points.
By end-2024, those gaps had compressed to 0.3, 1.6, and 3.7 points respectively. In February 2025, DeepSeek-R1 briefly matched the top US model outright. As of March 2026, Anthropic (Elo 1,503), xAI (1,495), Google (1,494), OpenAI (1,481), Alibaba (1,449), and DeepSeek (1,424) occupy the top tier of the Arena leaderboard, separated by 79 Elo points across the full cluster. The gap between the leading model and the weakest in that group is 2.7%.
The US still produces more notable models in absolute terms — 50 in 2025 versus China’s 30, per Stanford HAI. That 30 is a multiple of what China produced before 2022. The model count and the benchmark convergence together tell the same story: not that China has caught up on all dimensions, but that it has built a domestic AI research infrastructure capable of producing models that compete at the frontier.
China’s key flagship models as of mid-2026:
| Model | Organization | Architecture | Notable Capability |
|---|---|---|---|
| DeepSeek-R2 | DeepSeek AI | MoE + GRPO RL | Reasoning; flagship open-weights model globally |
| Qwen3-Max | Alibaba Cloud | MoE sparse attention | Multimodal; competes with GPT-5 tier |
| ERNIE 5.0 | Baidu | Dense transformer | Chinese language tasks; embedded in Baidu search |
| Doubao Pro | ByteDance | Proprietary | Consumer products; largest deployment by Chinese users |
| Kimi k2 | Moonshot AI | Sparse MoE | Long-context; 1M token window |
Sources: Model organization disclosures; Stanford HAI 2026 AI Index (Elo ratings as of March 2026)
Sarah Mitchell’s read on the model convergence: the benchmark number is real but requires context that the benchmark doesn’t supply. Arena Elo measures how often one model’s output is preferred over another’s in blind human evaluation — it is a useful relative measure that captures general capability. It does not capture deployment reliability at scale, adversarial robustness, or behavior under content constraints. DeepSeek-R2’s hallucination rate on a knowledge-vs-belief test dropped to 14.4% accuracy when a false claim was framed as a user belief, versus roughly 90% under normal testing (Stanford HAI 2026). The benchmark gap is 2.7%. The behavioral gap under adversarial conditions is substantially wider and substantially less reported.
China’s AI Patent Dominance — Volume vs. Impact
Patent volume is the most cited metric in discussions of Chinese AI innovation and the most frequently misunderstood. Understanding the actual data requires holding two true but incompatible-sounding facts simultaneously.
Fact 1: China holds 60% of global AI patent holdings and filed 69.7% of AI patents globally as of the Stanford HAI 2026 AI Index. WIPO’s World Intellectual Property Indicators 2025 Report confirms that Chinese inventors filed approximately 1.8 million total patent applications in 2024 — nearly half (49.1%) of the 3.7 million filed worldwide — and that China held 5.7 million patents in force, more than any other country.
Fact 2: Stanford HAI’s 2026 AI Index draws a consistent distinction between patent volume and patent impact. US AI patents attract substantially more citations than Chinese ones — a proxy for commercial and scientific influence that volume metrics don’t capture. The WIPO Patent Landscape Report on Generative AI (2024) found that Chinese inventors filed 38,210 generative AI inventions between 2014–2023, against 6,276 for US inventors — a 6:1 filing ratio. The citation ratio runs the other direction.
This is not a dismissal of Chinese AI innovation. It is a diagnosis: China’s IP strategy has historically optimized for volume, partly because the state incentive structure rewards filing count. That is changing. China’s 15th Five-Year Plan explicitly targets improvement in high-value international patent families, and the WIPO data shows China is the top PCT applicant origin for digital communication applications — a commercially significant category — with Huawei Technologies leading all PCT applicants globally for 2025.
The structural shift to watch is not whether China files more patents — it already files more than the rest of the world combined. The question is whether the citation profile of those patents converges toward the US profile over the next plan period. The 15th Five-Year Plan’s emphasis on basic research (basic research funding hit 7.08% of total R&D in 2025, crossing that threshold for the first time) suggests a deliberate policy push in that direction.
| Metric | China | United States | Source |
|---|---|---|---|
| AI patent holdings (% global) | 60% | — | Chinese Academy of Cyberspace Studies, Nov 2025 |
| AI patents filed (% of global total) | 69.7% | — | Stanford HAI 2026 AI Index |
| Total patent applications filed (2024) | 1.8M (49.1% global) | 501,831 | WIPO IP Facts and Figures 2025 |
| Patents in force (2024) | 5.7M (#1 globally) | 3.5M | WIPO IP Facts and Figures 2025 |
| PCT applications filed (2025) | #1 globally | #2 | WIPO Interactive IP Facts 2025 |
| Top PCT applicant organization | Huawei Technologies (7,523 apps, 2025) | Qualcomm | WIPO Interactive IP Facts 2025 |
Sources: Stanford HAI 2026 AI Index; WIPO World Intellectual Property Indicators 2025; WIPO Interactive IP Facts and Figures 2025; Chinese Academy of Cyberspace Studies, China Internet Development Report 2025 (released at 2025 WIC Wuzhen Summit, November 8, 2025)
China’s AI Research Output: Leading by Volume, Gaining in Impact
Stanford HAI’s 2026 AI Index establishes that China leads all countries in AI publication volume, with 23.2% of global AI publications and 20.6% of global AI research citations. The citation share is the number worth tracking: it suggests that Chinese AI research has moved from being voluminous-but-ignored to voluminous-and-read.
Greater China recorded a 27 percentage-point increase in organizational AI use in 2024 — one of the fastest growth rates globally in the Stanford HAI survey, which covers AI use in enterprises across regions. That adoption surge is both a demand signal and a research signal: Chinese labs have large-scale domestic deployment contexts that generate real-world feedback loops for model development.
China’s R&D spending provides the macro context. In 2025, total R&D investment reached 3.92 trillion yuan ($569 billion), representing 2.80% of GDP — a figure that exceeded the OECD average R&D intensity for the first time, per China’s Ministry of Science and Technology. Basic research funding specifically hit 280 billion yuan, or 7.08% of total R&D — crossing 7% for the first time in a signal that Beijing is pushing investment up the knowledge hierarchy toward the foundational research layer.
China’s AI Computing Infrastructure
Computing capacity is where China’s structural gap with the United States is most tangible and most policy-active.
China’s National Data Administration’s Digital China Development Report 2025 (released June 2026) reports that China’s intelligent computing power scale reached 1.59 million PFLOPS by the end of 2025, ranking second globally. The country operated more than 13.73 million standard server racks and had built 42 large-scale AI computing clusters — the 42 ten-thousand-GPU clusters mentioned by CNNIC’s 57th Report in a consistent cross-agency data point.
By comparison, China Academy of Information and Communications Technology (CAICT) data shows the United States accounted for approximately 32% of world total computing power of 910 EFLOPS as of end-2024, versus China’s 26%. RAND Corporation’s June 2025 analysis “Full Stack: China’s Evolving Industrial Policy for AI” puts the AI-specific compute share more starkly: China controls roughly 15% of total AI compute versus the United States at approximately 75% — a ratio that reflects not just data center count but the concentration of frontier AI accelerator chips, most of which are produced under US export control restrictions.
China’s stated policy target was 300 EFLOPS of total computing capacity by 2025, per the “Action Plan for High-Quality Development of Computing Power Infrastructure.” Reported figures — approximately 215 EFLOPS of AI-specific capacity, with 1.59 million PFLOPS intelligent computing — suggest China is tracking toward the target but has not closed the underlying chip-constraint gap, which is a design-and-fab dependency rather than a construction problem. By March 2026, MIIT data puts intelligent computing capacity at 1.882 million PFLOPS, up 2.5× year-over-year — infrastructure is scaling; the chip ceiling remains.
The “East Data, West Calculation” initiative, which routes data workloads to energy-rich western provinces, and the $295 billion national AI data center network plan announced in 2026 (per MIIT planning documents) are infrastructure responses to a hardware constraint that policy cannot directly resolve. China can build more data centers; it cannot without restriction buy more NVIDIA H100/H200 accelerators.
| Metric | Value | As of | Source |
|---|---|---|---|
| Intelligent computing capacity | 1.59M PFLOPS | Dec 2025 | National Data Administration, Digital China Development Report 2025 |
| Standard server racks in service | 13.73M | Dec 2025 | National Data Administration, Digital China Development Report 2025 |
| AI computing clusters (10k-GPU scale) | 42 | Dec 2025 | CNNIC 57th Report, Feb 2026 |
| Intelligent computing capacity (updated) | 1.882M PFLOPS | Mar 2026 | MIIT, via KuCoin infrastructure report |
| China share of world total computing | ~26% | End-2024 | CAICT, 2024 White Paper |
| US share of world total computing | ~32% | End-2024 | CAICT, 2024 White Paper |
| China AI-specific compute share | ~15% | 2025 est. | RAND Corp., Full Stack, June 2025 |
China’s AI Investment: The Private-Public Gap
The number most cited in AI geopolitics — the US–China private investment ratio — is 23.1:1. According to the Stanford HAI 2026 AI Index, US private AI investment reached $285.9 billion in 2025, against China’s $12.4 billion. That is not a contested figure. It is also a systematically incomplete one.
Stanford HAI’s report is explicit that “looking at just private investment figures likely understates China’s total AI spending, given its government guidance funds.” RAND Corporation’s June 2025 analysis of China’s industrial policy estimates that government guidance funds have deployed approximately $912 billion across industries since 2000, with AI and semiconductors constituting a primary focus of those deployments. In 2025, China announced a new 1 trillion yuan ($138 billion) state venture capital fund targeting AI, semiconductors, and advanced manufacturing — a figure larger than China’s entire reported private AI investment base over the past decade.
This does not close the gap. The US ecosystem advantage is not only capital; it is a dense private-sector innovation network, access to frontier compute, and a startup environment that can absorb capital more efficiently than state-directed programs. But the 23:1 ratio, taken as a headline measure of relative AI ambition, is misleading in precisely the way that a comparison of visible icebergs would be.
China’s corporate AI investment is concentrated in a small number of large technology companies. Alibaba announced ¥380 billion ($53 billion) in cloud and AI infrastructure investment over three years in early 2025. Tencent allocated ¥80 billion ($11.3 billion) specifically to AI in 2025 capital expenditure, per company disclosures. Huawei continues to invest heavily in domestic AI chip development (Ascend series) as a direct response to US export controls.
| Investment type | 2025 figure | Source |
|---|---|---|
| US private AI investment | $285.9B | Stanford HAI 2026 AI Index |
| China private AI investment | $12.4B | Stanford HAI 2026 AI Index |
| China core AI industry scale | >¥1.2T ($173.9B) | MIIT / Xinhua, March 2026 |
| China total R&D investment | ¥3.92T ($569B) | China NBS / MOST, March 2026 |
| Alibaba 3-year AI/cloud investment | ¥380B ($53B) | Alibaba corporate disclosures, 2025 |
China’s AI Policy Architecture
China’s approach to AI is not a set of regulations — it is an industrial policy stack that runs from semiconductor design through model development to application deployment and public adoption, with distinct layers of government involvement at each level.
The foundational policy document is the New Generation AI Development Plan (2017), which set the target of becoming the global AI leader by 2030. Subsequent policy layers — the 14th Five-Year Plan (2021–2025), the AI Plus strategy, and now the 15th Five-Year Plan (launched May 2026) — have progressively tightened the AI-industrial policy integration and expanded the list of sectors targeted for mandatory or incentivized AI adoption.
The regulatory layer that gets international attention is the algorithm and model registration system. As of March 2025, 346 generative AI services were registered with the Cyberspace Administration of China. Every service must comply with content requirements that include restrictions on training data and output topics deemed harmful to social stability. These requirements have two simultaneous effects: they make Chinese models more controllable domestically and less deployable internationally.
Manufacturing AI penetration — often undercovered in English-language analyses — is arguably the more strategically significant development. By the end of 2025, more than 30% of Chinese manufacturing enterprises with annual turnover exceeding ¥20 million had adopted AI technologies, per MIIT Minister Li Lecheng’s March 2026 NPC statement. MIIT’s lighthouse factory program had by that point achieved AI penetration exceeding 70% of business scenarios at designated sites, with over 6,000 vertical-domain industrial models deployed.
The 15th Five-Year Plan targets AI-physical integration as the primary economic growth driver for 2026–2030, with humanoid robotics as the headline application category. China had already released over 300 humanoid robot products by end-2025, per Li’s NPC statement. The robotics-AI linkage is where China’s manufacturing infrastructure advantage, its compute investment, and its model development converge into a single industrial application — and it is the quadrant of the China AI story that Western analysts most consistently underweight.
How Does China’s AI Talent Pipeline Compare?
Talent is the dimension where the China AI story becomes genuinely complex, and where the available data most conspicuously conflicts with the narrative of US structural dominance.
Stanford HAI’s 2026 AI Index documents two facts that sit in tension. First: the US produced more notable AI models, attracted more private investment, and hosts more data centers than any other country. Second: the number of AI researchers and developers moving to the United States has dropped 89% since 2017, with an 80% decline in the single year to March 2026 — a trend Stanford HAI attributes partly to H-1B visa restrictions under current US immigration policy.
China has built domestic AI talent at scale. The country graduated more STEM PhDs than the United States in every year from 2016 onward, by National Science Foundation and UNESCO data. China’s AI research publication volume — 23.2% of global output — is not produced by a small elite; it reflects a large research workforce embedded in universities, state research institutes, and corporate labs.
The constraint on Chinese AI talent is export restriction, not supply. Chinese AI researchers working on frontier hardware-constrained problems (large-scale training, custom silicon, post-CUDA stacks) are constrained by what chips they can legally access and what models they can legally train. DeepSeek’s R1 and subsequent work demonstrated that algorithmic innovation can, to a non-trivial degree, substitute for hardware — GRPO reinforcement learning allowed competitive model training at significantly lower compute cost. Whether that substitution holds as model scale increases is the unresolved empirical question.
China’s Generative AI Ecosystem: Domestic Models, Domestic Platforms
China’s generative AI ecosystem runs almost entirely on domestically developed models and platforms, because access to leading Western systems including ChatGPT is blocked on the mainland. This is not a competitive weakness — it is the condition under which a domestic AI industry was built from scratch at speed.
DeepSeek, Alibaba’s Qwen, ByteDance’s Doubao, Baidu’s ERNIE, and Moonshot AI’s Kimi collectively serve a user base that CNNIC measured at 602 million as of December 2025. Doubao holds the largest consumer user base among Chinese platforms; DeepSeek holds the largest international user base among Chinese-origin models and ranks among the top globally downloaded open-source models.
As of March 2025, 346 generative AI services were registered with the Cyberspace Administration of China — a larger count of regulated, licensed AI services than any other jurisdiction. Each registered service must meet content requirements; the registration database is effectively a complete census of the domestic market.
Chinese-developed open-source AI models ranked first globally in downloads over the preceding year, per MIIT Minister Li Lecheng’s March 2026 NPC statement. This is not trivially true: DeepSeek-V3 and its derivatives were downloaded at scale by developers globally despite (and partly because of) the US compute advantage these models demonstrated could be partially offset.
The structural pattern in Chinese AI deployment differs from the US pattern. CNNIC data shows 76% of Chinese generative AI users primarily use these tools for question-answering; 47.8% for image and video generation or processing; 10.8% for code generation. Industrial AI penetration — which MIIT tracks separately, not through CNNIC — shows a different picture: IDC data cited by China’s National Development and Reform Commission showed industrial AI penetration jumping from 9.6% to 47.5% in a single year. That is the number that matters for China’s manufacturing economy, and it receives far less international coverage than the consumer adoption story.
FAQ
What is China’s AI market size in 2026?
Multiple figures circulate, each correct under a different definition. China’s Ministry of Industry and Information Technology reported that China’s core AI industry — covering chips, computing infrastructure, models, and AI-enabled services — exceeded 1.2 trillion yuan ($173.9 billion) in 2025. Fortune Business Insights pegs the software and applications segment at $28.18 billion in 2025. Stanford HAI’s 2026 AI Index records $12.4 billion in private investment flows. Axis Intelligence Research recommends specifying the perimeter when citing any of these figures, because they measure genuinely different things.
How many generative AI users does China have?
As of December 2025, China had 602 million generative AI users, according to CNNIC’s 57th Statistical Report on Internet Development (February 2026). This represents 42.8% of China’s total population — a 141.7% year-on-year increase driven in large part by the January 2025 launch of DeepSeek-R1 and subsequent domestic platform deployments.
Has China caught up with the US in AI?
On model performance benchmarks, nearly. The US–China performance gap on the LMSYS Arena Elo leaderboard was 2.7% as of March 2026, per Stanford HAI’s 2026 AI Index — down from a 17–31 percentage-point range across key benchmarks in mid-2023. On private investment, no: the US spent 23 times more than China in private AI capital in 2025. On patent volume, China leads substantially. On computing infrastructure, China ranks second globally but holds roughly 26% of total world computing power against the US at 32%, with an additional constraint from export controls on frontier accelerator chips.
What AI models has China produced?
China’s most notable AI models as of mid-2026 include DeepSeek-R2 (reasoning, open-weights), Alibaba’s Qwen3-Max (multimodal), Baidu’s ERNIE 5.0 (Chinese language, search-embedded), ByteDance’s Doubao Pro (consumer), and Moonshot AI’s Kimi k2 (long-context). Stanford HAI’s 2026 AI Index records 30 notable AI models from China in 2025, against 50 from the United States. Chinese models including Alibaba and DeepSeek occupy the top tier of the Arena Elo leaderboard as of March 2026.
How much does China spend on AI research?
China’s total R&D investment reached 3.92 trillion yuan ($569 billion) in 2025, representing 2.80% of GDP — exceeding the OECD average for the first time, per the Ministry of Science and Technology. This figure covers all R&D sectors, not AI specifically. Private AI investment, measured by Stanford HAI, was $12.4 billion. The gap between total R&D and private AI investment reflects the state-directed, cross-sector nature of China’s innovation economy.
Does China lead in AI patents?
By volume, yes, substantially. China holds 60% of global AI patent holdings (Chinese Academy of Cyberspace Studies, November 2025) and filed 69.7% of all AI patents globally as of the Stanford HAI 2026 AI Index data. WIPO confirms China’s dominance across all patent categories: 1.8 million total applications in 2024 (49.1% of the global total), and 5.7 million patents in force — more than any other country. The volume lead does not translate equally to impact: US AI patents attract more citations, indicating higher commercial and scientific influence per patent. China’s 15th Five-Year Plan explicitly targets high-quality international patent development as a strategic objective.
What is the CAPI™ index?
The China AI Parity Index (CAPI™) is an original composite metric developed by Axis Intelligence Research to measure China’s convergence with the United States across five structural AI dimensions: model performance, research output, patent volume, computing infrastructure, and private investment. Each dimension is weighted by strategic importance and measurement reliability. The Q2 2026 baseline reading is 72.5 out of 100, with China scoring near parity on model performance (93.3) and patent volume (90.0), moderately on research output (80.0) and compute infrastructure (76.3), and very low on private investment (4.3). The index is updated quarterly at axis-intelligence.com/china-ai-statistics/.
What are China’s AI policy goals?
China’s national AI strategy targets global AI leadership by 2030, per the New Generation AI Development Plan (2017). The 15th Five-Year Plan (launched May 2026) places AI-physical integration — AI embedded in manufacturing, robotics, and industrial processes — as the primary economic growth driver for 2026–2030. Specific 2025 milestones achieved include more than 30% of qualifying manufacturing enterprises adopting AI (MIIT, March 2026) and over 300 humanoid robot products released. The regulatory layer includes mandatory registration of generative AI services with the Cyberspace Administration of China (346 registered as of March 2025) and content compliance requirements for all registered services.
Methodology
Axis Intelligence Research compiled this dataset through direct retrieval of primary-source documents between July 1–24, 2026. All statistics in this article correspond to a row in the downloadable CSV (CC BY 4.0).
Primary sources used: (1) Stanford HAI 2026 AI Index Report, accessed at hai.stanford.edu, April 2026, with supplementary Technical Performance chapter at hai.stanford.edu/ai-index/2026-ai-index-report/technical-performance; (2) CNNIC 57th Statistical Report on China’s Internet Development, February 5, 2026, cited via english.www.gov.cn; (3) China MIIT / Xinhua statement on core AI industry scale, March 5, 2026, at english.www.gov.cn; (4) China Ministry of Science and Technology / National Bureau of Statistics R&D communiqué, March 5, 2026, at english.www.gov.cn; (5) WIPO World Intellectual Property Indicators 2025, November 2025, at wipo.int/web-publications/world-intellectual-property-indicators-2025-highlights; (6) WIPO IP Facts and Figures 2025, at wipo.int/web-publications/ip-facts-and-figures-2025/en/patents-and-utility-models; (7) National Data Administration, Digital China Development Report 2025, June 2026; (8) RAND Corporation, “Full Stack: China’s Evolving Industrial Policy for AI,” June 2025, at rand.org; (9) Fortune Business Insights, China Artificial Intelligence Market Report 2025; (10) Chinese Academy of Cyberspace Studies, China Internet Development Report 2025, November 2025.
CAPI™ methodology: Five-dimension composite. D1 uses Arena Elo gap (2.7%) from Stanford HAI 2026 Technical Performance chapter; severity factor of 2.5 applied. D2 uses Stanford HAI publication-share data (23.2%) and notable model count ratio (30/50 = 60%); arithmetic mean. D3 uses Stanford HAI AI patent share (69.7%) with −10 quality adjustment. D4 uses CAICT compute share data (China 26%, US 32% of 910 EFLOPS) with −5 export control adjustment. D5 uses Stanford HAI private investment ratio ($12.4B/$285.9B = 4.3%). All inputs sourced from fetched documents, not memory. Baseline reading dated Q2 2026; next update expected Q3 2026 using NDA mid-year computing capacity data and CNNIC 58th Report.
Limitations: (1) Chinese AI market size figures span $12.4B–$173.9B depending on perimeter; all figures stated with their perimeter definition. (2) CAPI™ D5 (private investment) systematically understates China’s total AI capital mobilization by excluding state guidance fund deployment. The score would be materially higher if government AI capital flows were measured on a comparable basis. (3) Computing infrastructure figures from multiple sources use slightly different reference dates (end-2024 for CAICT, end-2025 for NDA); this article uses NDA for 2025 figures and CAICT for US comparison ratios. (4) Model performance benchmarks reflect Arena Elo as of March 2026; this score changes weekly. (5) CNNIC’s 602M user count uses a six-month active-use window.
About This Dataset
China’s AI statistics change materially every quarter. The CAPI™ index will be updated as Stanford HAI, CNNIC, NDA, and WIPO release new data. This article will be revised when: (a) CNNIC’s 58th Statistical Report is released (expected August 2026) with June 2026 user data; (b) Stanford HAI releases updated Arena Elo data or a mid-year supplement; (c) China’s NDA releases mid-year computing capacity figures; (d) any CAPI™ input dimension shifts by more than 5 points from the Q2 baseline.
Dataset download: [china-ai-statistics.csv] — CC BY 4.0
Citation (APA): Axis Intelligence Research. (2026). China AI Statistics 2026 [Dataset]. https://axis-intelligence.com/china-ai-statistics/. License: CC BY 4.0.
Citation (MLA): Axis Intelligence Research. “China AI Statistics 2026.” Axis Intelligence Research, July 2026, axis-intelligence.com/china-ai-statistics/. CC BY 4.0.
Citation (Chicago): Axis Intelligence Research. “China AI Statistics 2026.” Axis Intelligence Research, July 2026. https://axis-intelligence.com/china-ai-statistics/.
Corrections: If you identify an error in this dataset, please contact [email protected]. Verified corrections are posted with a dated correction note. See /corrections/ for our full corrections policy.
Others Pages
- AI Statistics: The Complete 2026 Data Guide
- AI Data Center Statistics 2026
- Robotics Statistics 2026