China’s 15th Five-Year Plan mentions AI over 50 times and embeds it across every sector of the national economy. Local governments are already subsidizing deployment ahead of regulatory consensus. The blueprint is the most ambitious state-directed AI strategy in history — but the gap between mandate and machinery has never been wider.
On March 5, 2026, China’s National People’s Congress opened with the release of the 15th Five-Year Plan — and a vocabulary shift that Western observers are only beginning to process. The 141-page document mentions artificial intelligence more than 50 times. It introduces a new planning term, 模芯云用 (model-chip-cloud-application), encoding a formal architectural decision: AI is not a sector to be regulated — it is the operating system for the entire economy.[1]
The word for lithography machine does not appear once. Neither does wafer fab or extreme ultraviolet. The entire vocabulary of the chip war as debated in Washington is absent from the document. What appears instead is a deployment strategy. The 70% semiconductor self-sufficiency target from Made in China 2025 — which China missed by roughly 50 points — has been quietly deleted and replaced with a deployment metric: digital economy value-added at 12.5% of GDP by 2030.[1][4]
This is not a catch-up plan. As Dan Wang of Eurasia Group observed, for the first time China explicitly wants to lead rather than follow.[5] The strategy spans AI, humanoid robotics, 6G, brain-computer interfaces, quantum computing, and nuclear fusion. State-owned enterprises are enrolled to create demand for domestic chips. R&D spending will increase by 7% over the next five years. And the “AI Plus” initiative — first launched as a campaign in August 2025 — is now embedded in the formal planning architecture that governs how China allocates capital, talent, and institutional attention.[2][3]
But the most revealing signal isn’t in the plan itself — it’s on the ground. Within days of the NPC opening, at least seven local governments rolled out subsidy programs to accelerate deployment of OpenClaw, an open-source AI agent platform. Shenzhen’s Longgang District offered up to 10 million yuan for companies building OpenClaw applications. Tencent hosted a free setup session that drew crowds ranging from children to retirees. ByteDance, Alibaba, Huawei, and JD.com all launched their own OpenClaw derivatives. Shenzhen’s municipal health commission ran a training session attended by thousands, under the banner of “AI Plus.”[6][7][8]
This is state-directed AI deployment at a scale and velocity that has no Western equivalent. The plan is real. The capital is flowing. The local governments are executing. The question is whether the machinery underneath — the software ecosystem, the compute infrastructure, the workforce readiness, and the products themselves — can bear the weight of the mandate.
China bets the house on AI. $200B Five-Year Plan. Tech self-reliance.
A DRIFT score of 50. A CUDA gap acknowledged in Chinese regulatory filings. Seven local governments subsidizing an Austrian developer’s open-source tool before the plan’s ink is dry. The blueprint outpaces the building.
“For the first time, China wants to lead in a number of technologies. Previously, the focus was always catching up with the West.”
— Dan Wang, China Director, Eurasia Group[5]China launches the AI+ campaign to accelerate integration of artificial intelligence across manufacturing, healthcare, education, and logistics through 2035. The first formal signal that AI will be embedded in national planning architecture.[3]
D4 Regulatory OriginDeepSeek proves that frontier AI capability is achievable with hardware constrained by US export controls. The model validates China’s strategy of deployment over fabrication.[1]
D5 Quality SignalThe 141-page plan mentions AI 50+ times, introduces 模芯云用 framework, sets 12.5% digital economy GDP target, and enrolls state-owned enterprises as domestic chip buyers. R&D spending increased 7%. Self-sufficiency target replaced with deployment metric.[1][2]
D4 → D3 → D6Longgang District publishes draft policy offering up to 10 million yuan in subsidies for OpenClaw applications, free computing resources, and discounted office space for “one-person companies.” Seven local governments follow within days.[6][8]
D1 Customer CascadeByteDance launches ArkClaw. Tencent hosts free setup sessions drawing thousands. Huawei launches Xiaoyi Claw on HarmonyOS. Alibaba unveils CoPaw. JD.com offers paid installation services. Baidu develops enterprise agents on ERNIE.[7][9]
D3 Revenue AccelerationThe plan is formally approved. Ministries, provincial governments, and state agencies begin translating the blueprint into implementation plans, funding programs, and technology procurement mandates.[4]
D6 OperationalChinese financial filings acknowledge NVIDIA’s ecosystem as the principal barrier to domestic AI chip substitution. Moore Threads’ IPO prospectus describes it as “not easily surpassable.” The Five-Year Plan leaves the software ecosystem gap conspicuously unmetriced.[1]
At Risk: D5 QualityThe cascade originates in D4 (Regulatory) — state-level industrial policy creating the conditions for everything else. The amplification flows through D3 (Revenue), D1 (Customer), D2 (Employee), and D6 (Operational). But D5 (Quality) and the software ecosystem gap carry the structural risk that could constrain the entire strategy.
| Dimension | Evidence |
|---|---|
| Regulatory (D4)Origin · 55 | 15th Five-Year Plan embeds AI across all sectors. “AI Plus” elevated to formal planning architecture. Deploy-first, regulate-later pattern established. SOEs enrolled as domestic chip buyers. R&D spending increased 7%. The state is the cascade accelerant — unlike market-driven adoption, this is a top-down mandate that bypasses the normal capital allocation cycle.[1][2] |
| Revenue / Financial (D3)L1 · 70 | 12.5% digital economy GDP target by 2030. Seven local governments deploying millions in AI subsidies within days. State-directed capital flows into computing infrastructure, AI agent deployment, and domestic chip procurement. State-subsidized competitors will undercut Western pricing in industrial AI applications. The capital is not aspirational — local governments are disbursing real subsidies to real developers within weeks.[4][6] |
| Customer (D1)L1 · 68 | Every industry vertical is now a mandated AI adopter under state policy. OpenClaw adoption events drew thousands. Tencent, ByteDance, Alibaba, Huawei, Baidu, JD.com, and Xiaomi all launched AI agent products within a single week. Nearly a thousand people queued outside Tencent’s Shenzhen HQ. Local governments are subsidizing faster than regulators can warn.[7][9] |
| Employee (D2)L1 · 58 | Plan includes workforce AI skills mandates. “One-person company” subsidies promote AI agents as digital employees. Health commission running AI training sessions for thousands. The tension: China is explicitly subsidizing a model where one founder plus AI agents replaces a team — while simultaneously promising employment generation. Unresolved in the plan’s own language.[3][6] |
| Operational (D6)L2 · 65 | National computing grid with resource pooling, unified scheduling, and crisis-override provisions. National Data Administration published standards. NDRC published implementation guidance. China is building compute sovereignty infrastructure — a national “reservoir” of strategic computing reserves with government-mandated minimum service guarantees during crises. This has no Western equivalent.[1] |
| Quality (D5)⚠ At Risk · 30 | NVIDIA’s CUDA ecosystem described as the “main obstacle” to AI chip substitution in Chinese financial filings. Moore Threads’ IPO prospectus calls it “not easily surpassable.” DeepSeek proved frontier capability with constrained hardware, but one model is not an ecosystem. The Five-Year Plan leaves the software ecosystem gap unmetriced — the only major strategic domain with aspiration but no measurement. State mandates don’t guarantee product quality.[1] |
“Teaching Mode” — the gap between China’s stated AI strategy (the most sophisticated state-level technology plan ever published) and its current execution capacity (CUDA dependency, nascent domestic AI ecosystem, unresolved employment/automation tension) is extreme. The plan is a 10; the machinery is a 3.5.
FETCH = Chirp (57.67) × DRIFT (50) × Confidence (0.80) = 2,307 → EXECUTE — HIGH PRIORITY
Score exceeds the 2,000 threshold for high-priority publication. Confidence at 0.80 reflects multiple primary sources including Reuters, CNN, The Diplomat, CNBC, Chatham House, and Chinese corporate filings.
-- The State Machine: 6D At-Risk Cascade
-- China AI+ National Strategy (15th Five-Year Plan)
FORAGE state_machine_cascade
WHERE type = "at-risk"
AND sector IN ["technology", "geopolitics"]
AND ai_mentions > 50
AND state_directed = true
ACROSS D4, D3, D1, D2, D6, D5
DEPTH 3
SURFACE cascade_map
DRIFT cascade_map
METHODOLOGY 85 -- most sophisticated state-level AI plan ever published
PERFORMANCE 35 -- CUDA dependency, nascent ecosystem, workforce tension
FETCH cascade_map
THRESHOLD 1000
ON EXECUTE CHIRP at_risk "State-directed AI deployment — 4/6 dimensions amplifying, CUDA gap constraining D5. Seven governments in seven days. The blueprint outpaces the building."
SURFACE analysis AS jsonRuntime: @stratiqx/cal-runtime · Spec v1.1: cal.cormorantforaging.dev · DOI: 10.5281/zenodo.18905193
The most analytically significant finding in the 15th Five-Year Plan is what’s absent. As The Diplomat documented in a detailed textual analysis, the entire vocabulary of the chip war — lithography machines, wafer fabs, EUV — has vanished from the planning document. In its place: a deployment architecture.[1]
The new planning term 模芯云用 encodes a formal architectural decision: the chip is one layer of four, alongside AI models, cloud infrastructure, and application deployment. This is not a concession on semiconductor self-sufficiency — it’s a strategic reframe. China is betting that controlling the application layer and the deployment infrastructure matters more than controlling fabrication, at least for the next five years.
The replacement of the 70% chip self-sufficiency target with a 12.5% digital economy GDP target is the clearest evidence of this pivot. China missed the fabrication target by roughly 50 points. Rather than doubling down on a losing metric, the plan quietly deletes it and substitutes a deployment metric that China can influence through exactly the kind of state-directed adoption happening on the ground.
“Beijing is trying to manage a ‘controlled glide’ in growth while building a new economy based on technology rather than property. It is a high-stakes rebalancing where the government is betting the house on AI and advanced manufacturing.”
— Andy Ji, Asian FX & Rates Analyst, ITC Markets[4]The speed of OpenClaw adoption in China is the single most revealing real-time indicator of the AI+ strategy’s execution velocity — and its contradictions.
OpenClaw is an open-source AI agent platform created by Austrian developer Peter Steinberger. Unlike chatbots, it connects large language models to real-world tools — operating desktops, calling APIs, managing files, executing multi-step workflows autonomously. It appeared on GitHub in November 2025 and went viral in China in early 2026.[7]
The response from Chinese institutions was extraordinary. Nearly a thousand people queued outside Tencent’s headquarters for free installation. Shenzhen’s health commission ran training sessions for thousands. JD.com launched a paid installation service. ByteDance created ArkClaw for browser-based deployment. Huawei built Xiaoyi Claw for HarmonyOS. Alibaba unveiled CoPaw. And within a single week, at least seven local governments — Shenzhen, Wuxi, Hefei, Changshu, Changzhou, Nanjing, and others — published subsidy programs offering millions of yuan for OpenClaw-based applications.[6][8][9]
The contradiction: Beijing’s state media published security warnings about OpenClaw’s access to personal data at the same time local governments were subsidizing its adoption. This deploy-first-regulate-later pattern — where execution velocity outpaces the regulatory framework — is the D4/D5 tension at the heart of the cascade. The plan creates the mandate. The local governments execute. The regulators trail behind.
For the global technology ecosystem, the OpenClaw wave demonstrates something important: China’s AI strategy is not about building frontier models. It’s about deploying AI agents at population scale, using whatever tools are available — including Western open-source frameworks — and building the institutional infrastructure around them. The plan says AI. The ground says OpenClaw. The gap between them is the DRIFT.
China deleted its failed 70% semiconductor self-sufficiency target and replaced it with a 12.5% digital economy GDP metric. The vocabulary shift signals a strategic pivot from fabrication to application — and Washington’s chip-war framing may be targeting the wrong variable.
China’s state-directed model deliberately creates DRIFT by mandating deployment before the execution machinery is ready. The assumption: deployment creates the forcing function that builds the machinery. If this works, the DRIFT closes. If it doesn’t, the state has subsidized adoption of tools it can’t control.
The Five-Year Plan metricizes everything — GDP share, R&D spending, compute clusters. The one major domain with aspiration but no measurement is the software ecosystem. Chinese companies acknowledge in their own filings that NVIDIA’s ecosystem remains the principal barrier. The plan’s silence is the most important thing it doesn’t say.
The velocity of local government response to the AI+ mandate is the real execution signal. It suggests China’s bureaucratic machinery is calibrated for speed on AI deployment in a way that has no Western democratic equivalent. Whether this speed produces quality or waste will define the next five years.
UC-066 maps the demand side of the AI infrastructure buildout. UC-065 maps the supply side. The CUDA gap that constrains China’s D5 is the same moat that powers Nvidia’s dominance.
UC-065 The Treadmill — Nvidia’s CUDA ecosystem is the gap China can’t close · UC-064 The Great Swap — Meta’s $162B flows to the infrastructure China is trying to replicate · UC-059 The Code Is Dead — AI agents run on CUDA; OpenClaw runs on CUDA · UC-047 The 21-Mile Chokepoint — geopolitical supply chain dependency
One conversation. We’ll tell you if the six-dimensional view adds something your current tools miss — or confirm they have it covered.