China's AI Strategy: The Race to 2035

China is executing a whole-of-nation AI mobilization through a three-pillar incentive architecture — state investment that funds supply, subsidies that lower costs, and regulatory mandates that guarantee demand — producing deployment speed and scale no other country can match. But the quality gaps, the compute reality, and the commercialization paradox reveal both the power and the limits of this model.

The Strategy

China’s AI ambition isn’t a recent pivot. It’s a phased, whole-of-nation program that began with the 2017 AI Development Plan and extends to 2035, with a military-civilian fusion timeline reaching 2049. The strategy has evolved through three policy generations, each building on the last.

The 2017 plan set the foundation: make China the world’s primary AI innovation center by 2030, with a core AI industry target of 150 billion yuan by 2025. By 2023, the actual figure hit 578.7 billion yuan — four times the target, two years early. By late 2025, it exceeded 1 trillion yuan. Whatever else you think about the Chinese model, the targets aren’t aspirational. They’re being hit.

The 15th Five-Year Plan, previewed in October 2025, shifts emphasis from breakthroughs to application, conversion, and ecosystem building. Its core themes — scientific self-reliance, “New Quality Productive Forces,” and high-quality development — signal a pivot from building AI vertically (labs, models, chips) to diffusing it horizontally across science, industry, consumption, governance, and global cooperation.

What makes this unique isn’t any single target. It’s the institutional machinery. The whole-of-nation model aligns the State Council, three co-equal lead agencies (NDRC, MOST, MIIT), the PLA, universities, state-owned enterprises, and the private sector around shared objectives. In January 2026, all four agencies jointly issued the first national framework for government investment fund governance — distinguishing between strategic chokepoints (national funds) and regional strengths (provincial funds). The military dimension runs in parallel: mechanization-informatization-intelligentization integration by 2027, intelligent combat systems by 2035, world-class military status by 2049.

The National Champions

China’s AI ecosystem spans the full stack, and the companies at its center are solving different problems.

DeepSeek proved that algorithmic efficiency can partially offset hardware constraints. Its V3.2 model trained for just $5.57 million while scoring gold-medal level on AIME — a demonstration that brute-force compute isn’t the only path to frontier performance. Alibaba’s Qwen pushes multilingual reach across 119 languages and open-source distribution. Moonshot’s Kimi specializes in long-context processing — 2 million Chinese characters — and raised $500 million at a $4.3 billion valuation. Zhipu’s GLM-Image is perhaps the most strategically significant: the first major open-source model trained entirely on Huawei Ascend chips, with no Nvidia hardware and no CUDA dependency, demonstrating that the domestic hardware stack is viable for at least some workloads.

The open-source strategy is deliberate and increasingly effective. Chinese models now command 17.1% of global downloads, up from 1.2% at the end of 2024, with peaks reaching 30%. This isn’t altruism — it’s a competitive weapon. Openness builds global developer dependency on Chinese models, contrasting sharply with the American approach of API-only, cloud-locked distribution.

The Talent Pipeline

The numbers here are staggering. China produces roughly 53,400 science and engineering doctorates annually — 1.5 times the US figure, with 37% in engineering. Four Chinese universities now rank in the global top 10 for AI conference publications: Tsinghua at number two (surpassing MIT and Stanford), followed by Peking, Zhejiang, and Shanghai Jiao Tong. In 2024, Chinese researchers published 23,695 AI papers — 42% of global output, matching the combined production of the US, UK, and EU-27.

The talent machine extends beyond organic production. The Qiming Plan recruits overseas PhDs under 40 with up to 5 million yuan in salary subsidies, 300,000 yuan in research funding, and six-figure living allowances. “Bounty-as-a-Service” models offer up to $700,000 for specialized chip and AI-adjacent engineers. AI job openings grew 543% year-over-year in 2025.

The Incentive Machine

This is the centerpiece. China’s AI incentive system isn’t just subsidies or just investment — it’s an integrated machine with three interlocking pillars, each reinforcing the others. This is the operational mechanism of the whole-of-nation model, and it has no Western equivalent.

Pillar 1: Funding the Supply Side. State capital at a scale no other country matches. An $8.3 billion national AI fund launched in March 2025. A $140 billion government-backed fund for AI and emerging tech over 20 years. The National IC Fund Phase III — $47.5 billion for chips, with 70% directed at foundational technology. Policy alignment is weighted at 60% in fund evaluations; financial returns are explicitly secondary.

For context: China’s private AI investment in 2024 was $9.3 billion versus America’s $109.1 billion. Cumulatively from 2013 to 2024, it’s $119 billion versus $470 billion. The state capital isn’t replacing private investment — it’s supplementing a shortfall.

Pillar 2: Lowering Costs at Every Layer. The subsidy architecture is remarkably granular. Corporate tax drops to 15% for High/New Tech Enterprises versus the standard 25%. R&D gets a 175% super deduction. Compute vouchers operate across 17-plus provinces, ranging from $140,000 to $1.1 million per company — Shanghai alone runs a 600 million yuan program covering up to 80% of AI compute rental costs. Data centers receive up to 50% electricity subsidies, conditional on using domestic AI chips. Land pricing is set below market at 70% minimum for encouraged industries. And perhaps most remarkably, AI education is now mandatory: Tianjin requires compulsory weekly AI classes, Beijing has deployed AI curricula across 1,400 schools reaching 1.83 million students.

Pillar 3: Guaranteeing Demand. This is the pillar the West doesn’t have. MIIT mandates 70% AI penetration in intelligent terminals by 2027 — mandatory for state-owned enterprises and local governments, rising to 90% by 2030. The energy-subsidy gating ties Pillar 2 directly to Pillar 3: you get cheap power, but only if you buy Chinese chips. Mandated adoption creates a guaranteed market, de-risks further investment, and attracts state and private capital back into the ecosystem. The three pillars don’t just coexist — they compound.

The EU offers programmatic funding — Horizon Europe, the InvestAI Facility aiming to mobilize €200 billion — but has no compute vouchers, no energy subsidies for AI, no mandatory AI education, no state-directed talent recruitment, no compliance targets. The EU has Pillar 1, partially. The US has even less on the incentive front. American AI leadership rests on private capital, market dynamics, and the sheer scale of its tech ecosystem — a market-pull model versus China’s state push.

The Compute Gap

The hardware reality is stark. China holds roughly 110,000 H100-equivalent GPUs versus America’s 850,000 and the EU’s 50,000. In global AI supercomputer performance share, it’s 14.1% versus 74.5% versus 4.8%.

Huawei’s Ascend 910C is the centerpiece of the domestic chip effort, but the performance claims don’t survive scrutiny. Huawei’s marketing says 8-12% behind the H100. DeepSeek’s researchers put it at roughly 60% of the H100. Against the Nvidia H200, it manages 76%. Against the GB200 for inference, just 20-30% per chip. Huawei’s system-level workaround — the CloudMatrix 384, clustering 384 Ascend 910C chips — approaches a single GB200 NVL72 system, but uses 5.3 times more chips to get there.

The real-world experience is mixed. DeepSeek initially tried Huawei chips for their flagship model but found the results unacceptable, switching to Nvidia H20 GPUs. Zhipu’s GLM-Image, by contrast, trained successfully on the Ascend stack. The domestic chips work for some workloads but not yet for frontier training.

The 2027 chip independence target — Beijing’s stated goal of AI hardware independence — faces long odds. Current realities include 75% Nvidia dependence, the 60% performance gap, and sub-30% yields at SMIC. The target will likely be redefined rather than achieved.

The Patent Paradox

China dominates AI patent volume at a staggering scale — nearly 70% of global granted patents in 2023, a 4.4-to-1 filing ratio over the US, and six times more generative AI inventions cumulatively. But citation rates tell a different story entirely. US patents receive an average of 13.18 citations each versus China’s 1.90 — a seven-times quality gap suggesting many Chinese patents have limited downstream impact. Germany, for comparison, achieves a citation rate of 6.12, more than triple China’s despite marginal volume.

The question is whether China’s citation gap is a temporary lag — citations take years to accumulate, and China’s patent surge is recent — or a structural difference in research impact. If the gap is closing, China’s volume advantage becomes overwhelming. If it’s structural, quantity without quality may not translate to technological leadership.

The Deployment Paradox

China’s deployment picture contains a striking contradiction. Chinese enterprises lead in agentic AI adoption — 46% piloting or deploying versus 40% in the US and 30% in the EU. But at the population level, AI usage runs lower: roughly 25% versus 28.3% in the US and as high as 46.4% in Norway.

The deployment strength is industrial, not digital. China installs over 50% of the world’s industrial robots. Automated ports provide the most concrete evidence of what the model produces: Zhoukou Port runs 12 automated cranes processing over a million containers with 30% productivity gains. Qingdao Port uses vacuum-based automated mooring that reduced berthing from 20-30 minutes to under 30 seconds. Tianjin Port operates fully unmanned with Beidou navigation, 5G, and AI robots.

But the commercialization gap remains jarring. Chinese platforms account for 46% of global AI monthly active users yet capture just 1.23% of top 100 AI company revenue. Revenue splits 89% enterprise, 11% consumer. The open-source strategy builds adoption and developer dependency — but it complicates monetization.

Case Study: OCR

OCR offers the cleanest illustration of the deployment thesis. In 2024, the best models hit roughly 75% accuracy on complex documents. Western vision-language models improved this to 77-88% through the first half of 2025. Then Chinese labs took over.

By late 2025, new models were pushing past 83%. In January 2026, the field exploded: DeepSeek-OCR-2 hit 91.09%, and Baidu’s PaddleOCR-VL-1.5 reached 94.5% — an 11.5 percentage-point improvement in a single month, more than the entire previous year combined.

This wasn’t just talent or funding. It was the combination of state compute subsidies lowering the cost of experimentation, open-source culture enabling rapid iteration and knowledge sharing, and a massive domestic market for document processing creating both training data and market pull. For printed documents, OCR is now solved infrastructure. For the whole-of-nation thesis, it’s proof of concept.

What’s Working, What’s Not

The strengths are real: academic excellence with four universities in the global top 10, algorithmic innovation proving efficiency can offset hardware gaps, open-source reach growing from negligible to 17.1% of global downloads, a talent pipeline producing 53,400-plus doctorates annually with aggressive overseas recruitment, coordinated multi-agency governance, regional specialization with provinces adapting AI to local strengths, and a multi-layered subsidy architecture unlike anything in the West.

The weaknesses are equally real: a monetization gap where 46% of global users produce 1.23% of revenue, a per-chip hardware gap with Ascend at roughly 60% of the H100 and DeepSeek unable to use it for frontier work, a patent impact deficit with a seven-times citation gap versus the US, contradictory enterprise adoption data, and constrained production capacity with SMIC yields below 30%.

China is building AI the way it builds highways — through state mobilization, mandated adoption, and subsidized scale. The model produces extraordinary deployment speed. Whether it produces the kind of deep, commercializable, quality-driven innovation that defines long-term technological leadership remains the open question.