Somewhere in a Beijing conference room, someone circled "artificial intelligence" on a whiteboard so many times the marker ran dry. China's 15th Five-Year Plan (2026-2030) just landed, and it reads less like a policy document and more like a love letter to AI - complete with a $62 billion science budget, a 10% bump from last year, and enough ambition to make Silicon Valley double-check its stock portfolio (You, 2026).
The "AI Plus" Program (Yes, That's Its Real Name)
China's "AI Plus" action plan, released in August 2025, sets targets that sound like they were generated by an overly optimistic AI model: 70% AI penetration across industries by 2027, 90% by 2030, and "ubiquitous deployment" by 2035 (MERICS, 2026). That's right - by 2035, your toaster in Shanghai will probably run a language model. Analyst Rebecca Arcesati diplomatically notes these function as "signals rather than actual targets," which is the policy-wonk equivalent of saying "let's see what sticks."
But here's what makes this more than a wishlist stapled to a press release: China is putting serious money where its mouth is. We're talking R&D spending increasing at least 7% annually through 2030, with priority funding for national laboratories and breakthrough projects in quantum computing, robotics, 6G, brain-computer interfaces, and embodied AI (Nature, 2025).
The Chip Problem (or: You Can't Run AI on Good Intentions)
Every ambitious AI plan eventually runs into the same brick wall: semiconductors. China needs advanced chips to train its models, and thanks to US export restrictions, getting them has become about as easy as returning something to IKEA without a receipt.
The response? Throw approximately €60 billion in subsidies at domestic chipmakers. Huawei is ramping up production of its Ascend 910C AI chips, targeting 1.6 million dies by 2026 (IEEE ComSoc, 2025). SMIC, China's top foundry, plans to expand advanced-node capacity to 80,000 wafers per month by 2027 (Tom's Hardware, 2025). The catch? SMIC is doing all this without access to ASML's extreme ultraviolet lithography machines - basically trying to paint the Mona Lisa with a house brush. Impressive effort, questionable brushwork.
DeepSeek Changed the Math
If one thing gave Beijing's techno-optimism a shot of adrenaline, it was DeepSeek. When DeepSeek-R1 dropped in January 2025 - an open-source model with 671 billion parameters built for under $6 million that briefly knocked ChatGPT off the App Store's top spot - it wasn't just a technical achievement. It was proof of concept for China's entire "do more with less" AI philosophy (Stanford HAI, 2025).
Chinese open-source models have since surged from about 1% of global usage in late 2024 to roughly 30% by mid-2025, with Alibaba's Qwen family overtaking Meta's Llama in total Hugging Face downloads (SCMP, 2025). Chinese models also cost one-fourth to one-sixth what their American counterparts charge, which turns out to be a pretty effective growth strategy.
The Governance Play Nobody Expected
While Washington has been leaning into deregulation, Beijing made a move that caught observers off-guard: proposing a World Artificial Intelligence Cooperation Organization (WAICO), pitched by Premier Li Qiang at the 2025 World AI Conference in Shanghai (CNN, 2025). The idea is essentially a UN for AI - complete with a technology-sharing platform and an algorithmic compensation fund financed by a royalty on commercial AI revenues.
Whether WAICO gets traction or becomes another acronym collecting dust is an open question. But the strategy behind it is clear: if you can't beat them on chips, beat them on rules. China is positioning AI as an "international public good" and targeting buy-in from the Global South - countries that would very much like a seat at the AI table without paying Nvidia prices for admission (Nature, 2025).
What Actually Happens Next
The honest answer: nobody knows. China's AI ambitions are a cocktail of genuine technical progress (DeepSeek, Qwen), brute-force investment (that $62 billion budget), and targets that may or may not survive contact with reality. The semiconductor bottleneck is real. The talent pipeline is growing but strained. And "ubiquitous AI by 2035" is the kind of promise that ages either like wine or like milk.
What's clear is that the AI landscape is no longer a one-country show. If you're building tools that work with AI-generated content - whether that's mapping complex reasoning chains with something like mapb2.io or auditing AI-driven web experiences - the models powering them increasingly speak Mandarin.
The old framing of a single AI "race" with a winner and loser might be the wrong metaphor entirely. As Foreign Affairs put it, the more likely outcome is "asymmetric AI bipolarity" (Foreign Affairs, 2025) - two very different approaches to AI, both reshaping the world, neither going away. The future of AI might not be American or Chinese. It might just be... both.
References
-
You, X. (2026). China intensifies push to become world leader in tech and AI. Nature. https://doi.org/10.1038/d41586-026-00814-3
-
You, X. (2025). China seeks self-reliance in science in next five-year plan. Nature. https://www.nature.com/articles/d41586-025-03491-w
-
Nature Editorial. (2025). China is leading the world on AI governance: other countries must engage. Nature. https://www.nature.com/articles/d41586-025-03972-y
-
Arcesati, R. (2026). China's next five-year bet on AI: Self-reliance, diffusion, and a lot of hype. MERICS. https://merics.org/en/comment/chinas-next-five-year-bet-ai-self-reliance-diffusion-and-lot-hype
-
Stanford HAI. (2025). Beyond DeepSeek: China's Diverse Open-Weight AI Ecosystem and Its Policy Implications. https://hai.stanford.edu/policy/beyond-deepseek-chinas-diverse-open-weight-ai-ecosystem-and-its-policy-implications
-
Tse, E. (2025). The Myth of the AI Race. Foreign Affairs. https://www.foreignaffairs.com/united-states/myth-ai-race
Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.