In March 2026, the Chengdu city government released its "AI+" action plan, targeting an AI industry scale of 2.6 trillion yuan by 2027 and a smart terminal penetration rate exceeding 70%. At first glance, this is another local industrial policy from China—ambitious, top-down, and heavy on buzzwords. But for those of us who track the intersection of code and chaos, a different signal emerges. This plan, if executed with even 60% fidelity, will generate a demand for compute, storage, and data services that far exceeds the capacity of centralized cloud providers in western China. And where centralized infrastructure fractures, decentralized networks find their load-bearing moment.
The narrative of AI dominance has always been about models—GPT, Gemini, Llama. But in 2026, the truer story lies beneath the algorithms: the infrastructure layer. Chengdu's plan, while silent on blockchain or crypto, inadvertently maps directly onto the thesis I developed during the 2024-2026 AI-Agent economic cycle: autonomous agents require decentralized identity, verifiable compute, and micropayment rails. This city of 21 million, with its electronics manufacturing backbone and government-controlled pilot projects, could become the proving ground for a new class of DePIN (Decentralized Physical Infrastructure Networks) tokens.
Context: The Plan in Numbers
Let me translate the policy from Mandarin to metrics. The 2.6 trillion yuan figure (approximately $360 billion at current exchange rates) represents a compound annual growth rate of over 30%—twice the national average for AI-related industries. To put that in perspective, the entire global AI chip market in 2025 was estimated at $150 billion. Chengdu is essentially aiming to build a $360 billion AI economy from a standing start in three years.
The plan's operational core is the "Double 100" program: 100 innovative AI products and 100 demonstration scenarios, with 20 benchmark scenarios funded each year. These scenarios target healthcare (West China Hospital), finance (Chengdu Bank), education (Sichuan University), and manufacturing (Foxconn Chengdu). The language echoes every other Chinese industrial policy—but the scale and specificity of the penetration target (>70% of smart terminals by 2027) forces a hard technical constraint.
What is a "smart terminal" in Chengdu's lexicon? Based on the city's electronics supply chain—Foxconn assembles iPhones, Huawei devices, and smart home products here—these are consumer-grade AI-enabled hardware. But the policy also mentions "agents," a term that, in the crypto world, has evolved to mean autonomous software entities executing on behalf of users. This is where the infrastructure requirement becomes non-trivial.
Core: The Hidden Compute Demand and DePIN's Opening
During the 2020 DeFi summer, I learned that every explosion in application layer activity is preceded by a bottleneck at the infrastructure layer. The same is true for AI. Chengdu's 2600 billion yuan target implies a massive increase in both training and inference compute. Based on my audits of AI infrastructure projects, I estimate that a city-wide penetration rate of 70% for smart terminals would require approximately 100 exaFLOPs of total inference compute by 2027. Chengdu's current capacity—the National Supercomputing Center (100 petaFLOPs) and the Tianfu Intelligent Computing Center (targeting 1 exaFLOP by 2025)—is insufficient by a factor of 100.
This is where decentralized compute networks enter the picture. Projects like io.net, Akash Network, and Render Network tokenize idle GPU capacity globally. Their combined available compute as of Q1 2026 is roughly 50 exaFLOPs—half of what Chengdu alone will need. The narrative that DePIN is only for speculative crypto mining is false; these networks are now attracting real-world workloads from AI startups. What they lack, however, is the kind of guaranteed, subsidized demand that a Chinese city can provide.
Which brings me to a concrete investment thesis: any protocol that can secure a pilot project within Chengdu's "Double 100" framework will see its token value re-rate significantly. The city government may not explicitly endorse crypto, but it will procure compute services. If those services come from a decentralized network operating through a legal intermediary (a common structure in China), the underlying token becomes a proxy for municipal AI infrastructure spending.
The Data Labeling Bottleneck
Chengdu's plan also calls for "700+ enterprise-level AI application scenarios." Each scenario will require high-quality, domain-specific training data. The city has a cost advantage in labor—AI training data annotators earn roughly $8,000 per year in Chengdu versus $15,000 in Beijing. This is a classic opportunity for a tokenized data marketplace: projects like Grass (which rewards users for scraping and labeling web data) or Synesis (a decentralized annotation protocol) could partner with local universities to create China-compliant annotation networks. The contrarian insight here is that data labeling, not compute, will be the first bottleneck for Chengdu's AI push.
Contrarian: The Risks That the Plan Ignores
No policy analysis is complete without auditing the narrative. Chengdu's plan, like most Chinese industrial policies, is silent on three critical risks: goal inflation, compute sovereignty, and talent retention. Let me stress-test each.
First, the 2600 billion yuan figure. Based on my experience auditing smart contracts for DeFi protocols, I know that numbers without transparent methodology are artifacts of optimism. The plan does not disclose baseline definitions. Does "AI industry scale" include the incremental value of traditional products enhanced with AI features? If so, then a smartphone with an AI camera chip counts fully toward the target, even if the AI component is a $5 SoC. In my audit reports, I always flag such non-standard revenue recognition. The same applies here: investors should demand a breakdown of pure AI revenue versus AI-adjacent revenue.
Second, compute sovereignty. Chengdu's supercomputers rely on chips from NVIDIA (H100 and B200) and domestic alternatives from Huawei (Ascend 910). The US export controls on advanced AI chips to China remain in place as of 2026. If Huawei's supply chain faces disruption, Chengdu's compute expansion could stall. Decentralized networks, ironically, offer a hedge: protocols like Akash aggregate GPUs from non-sanctioned regions (e.g., Europe, Southeast Asia), providing a sovereign compute layer that bypasses chip embargoes. But China's regulatory stance on crypto payments makes this channel legally murky.
Third, talent cost. The analysis notes that AI engineer salaries in Chengdu have risen to near Tier-2 peaks. In a bull market for AI, the city may find itself subsidizing headcount rather than innovation. The true measure of success is not the number of graduates but the number of net-new AI startups that achieve profitability without subsidies. My 2021 BAYC analysis taught me that community value survives only when speculation gives way to utility. The same holds for government-subsidized AI ecosystems: if the subsidy stops, the utility must persist.
Where the Architecture of Trust Rebuilds
On the ethics dimension, the plan is conspicuously bare. No mention of algorithmic auditing, data privacy, or liability for AI-caused harm. This is dangerous but predictable. In 2022, after Terra's collapse, I launched "The Solvency Audit" series, tracing contagion risks through dependent protocols. A similar audit is needed here: each of the 100 demonstration scenarios should have an independent risk assessment for bias, security, and failure modes. Blockchain-based audit trails—where every inference request and decision is hashed and timestamped—could provide the transparency that the plan lacks. Projects like OriginTrail or Lit Protocol are already building the infrastructure for verifiable AI outputs.
Takeaway: The Next Narrative
The failure mode of Chengdu's plan is a centralized, state-directed AI market that squeezes out innovation and leaves the city with a hollowed-out tech sector. The success mode is a thriving ecosystem of decentralized infrastructure providers, tokenized data markets, and on-chain AI governance pilots. As a narrative hunter, I am watching for three signals: (1) whether any DePIN project announces a partnership with a Chengdu-based enterprise; (2) whether the city issues a "compute voucher" program that can be programmatically executed on a smart contract; and (3) whether the first IPO from the program involves a company that integrates token incentives.
Composability is the new currency of innovation. Chengdu's plan is a massive, unwitting bet on the thesis that AI infrastructure must be decentralized to scale cost-effectively. The architecture of trust, rebuilt line by line, will determine whether this $360 billion dream becomes a load-bearing pillar of the next market cycle or a fracture in China's technology roadmap.
Auditing the narrative, not just the numbers.
— Where code meets chaos, truth emerges. — The architecture of trust, rebuilt line by line. — Culture codes the value; we just decode it.
