Hook: The 2.6 Trillion Yuan Blind Spot
Chengdu’s newly released “AI+” Action Plan targets a staggering 2.6 trillion yuan ($360 billion) in AI-related industrial output by 2030. The numbers are ambitious: a 70% penetration rate for “next-generation intelligent terminals and agents” by 2027, reaching 90% by 2030. But for anyone who has audited smart contracts or traded on on-chain order books, the silence is deafening. Not a single mention of blockchain, distributed ledger, or decentralized infrastructure. This isn’t just an oversight—it’s a strategic blind spot that could undermine the entire plan.
I’ve watched liquidity fragment across DeFi protocols. I’ve seen code-is-law promises break under reentrancy attacks. And I’ve learned that when a government plan ignores the foundational layer of trust and settlement, it’s building on sand. Chengdu’s plan reads like a 2018 whitepaper: heavy on vision, light on execution mechanics. The missing piece isn’t a feature—it’s the operating system for verifiable, transparent AI markets.
Context: The City That Powers Bitcoin Mining
Sichuan province, where Chengdu sits, has historically been China’s bitcoin mining heartland. Cheap hydropower from the summer monsoon made it the #1 destination for miners before China’s 2021 crackdown. Post-ban, mining moved underground or overseas, but the infrastructure remains. The Tianfu Supercomputing Center (1000 petaflops planned) and the National Supercomputing Center in Chengdu are now being marketed for AI workloads. The city’s low electricity costs—around $0.03–0.05/kWh—are still a competitive advantage.
The action plan targets 20 benchmark application scenarios annually, with “100 innovation products” and “100 demonstration scenarios” over the next few years. Priority sectors include electronics manufacturing, automotive, finance, and cultural tourism. These are all industries where on-chain data provenance, supply chain tracking, and tokenized assets could unlock massive efficiency gains. Yet the policy framework remains firmly centralized: government procurement, subsidies, and state-backed funds (rumored to be a 10-billion yuan AI industry fund). No mention of using smart contracts for transparent distribution of subsidies, no tokenization of data rights, no decentralized compute networks.
Core: The Order Flow Analysis of a Centralized AI Plan
Let’s break down the numbers through a battle trader’s lens. The plan claims 30%+ annual growth for the AI sector. That growth requires order flow—real demand from enterprises and consumers. But what kind of demand?
- The 70% penetration target: If this refers to device-level AI (AI smartphones, AI PCs), it’s largely driven by consumer upgrade cycles, not policy. But if it means B2B adoption in factories and government departments, then the “order flow” is essentially government spending. In DeFi terms, that’s like a liquidity pool with a single large LP (the state). When that LP withdraws—due to budget cuts or political shifts—the pool dries up.
- The 2600 billion yuan target: Based on my experience modeling yield farm returns, I suspect this figure includes a massive amount of “attributed” revenue—traditional products with an AI sticker slapped on. In crypto, we call this “vapor yield.” The real addressable market for pure AI services (model APIs, training, inference) is likely <30% of that number. Without independent on-chain metrics or auditable data, investors are left trusting the narrative, not the code.
- The infrastructure bottleneck: The Tianfu Supercomputing Center is designed for high-performance computing, not distributed inference. Training a 70B-parameter LLM requires tens of thousands of GPUs. Chengdu’s plan doesn’t specify chip sources. Given US export controls on NVIDIA H100s, the city will likely rely on Huawei’s Ascend chips. But Huawei’s ecosystem is walled garden—less composable than CUDA. This creates vendor lock-in, similar to being stuck on a single blockchain with high gas fees and no interoperability.
From a risk management perspective, the plan lacks a hedge. The entire AI economy in Chengdu is built on centralized compute, centralized data, and centralized governance. One data center outage, one regulatory change, one chip supply disruption—and the whole stack goes down. In crypto, we diversify across rollups and L2s. Here, there is no redundancy.
Contrarian: The Real Opportunities Lie in the Blind Spots
The conventional take is that Chengdu is doubling down on AI to compete with Beijing (research), Shenzhen (hardware), and Hangzhou (cloud). My contrarian view: the biggest alpha is in the areas the plan ignores.
Decentralized compute networks like Akash, Render, and io.net can provide cheaper, more resilient inference for the 70% terminal penetration target. Chengdu’s cheap hydropower—historically used for Bitcoin mining—could be repurposed for these networks. Instead of building state-owned data centers, the city could lease capacity from a global network of GPU providers, lowering cost and increasing fault tolerance. The plan’s silence on blockchain means this opportunity is being left to other cities, perhaps Shaoxing or Shenzhen, which are already experimenting with decentralized AI.
Data provenance and ethical AI: The plan completely omits AI safety and ethics. In China, the Generative AI regulation requires content filtering and algorithm filing. But without a transparent audit trail—powered by blockchain—how can citizens verify that AI decisions in healthcare or finance are free from bias? A smart contract-based registry of training data and model versions would increase trust without increasing censorship. The EU AI Act already pushes for such transparency; Chengdu could lead by grafting a public ledger onto its AI infrastructure.
Tokenized incentives for data labeling: The plan hopes to spawn 700+ enterprises using AI. Those enterprises need high-quality labeled data. Rather than centralizing data tasks in a government-owned facility, the city could issue tokens to compensate individuals for contributing data (privacy-preserving, of course). This model has been proven by projects like Grass and Nym. It turns every citizen into a stakeholder, driving adoption without massive upfront subsidies.
The contrarian move is not to compete on raw compute or model size—it’s to build a decentralized layer beneath the centralized AI stack. That’s where the efficiency gains compound.
Takeaway: The Real Test Will Be On-Chain
Markets don’t trust plans; they trust data. The only data that survives a bear market is on-chain proof of usage, liquidity, and real yields. When the first batch of “100 innovation products” is announced, I will be looking for verifiable on-chain metrics: number of transactions, unique active wallets, total value locked in AI-related smart contracts. If the city issues an AI industry bond, I’ll check whether it’s tokenized. If a local company claims to power the government’s AI systems, I’ll verify with a block explorer.
Chengdu has the energy, the talent, and the ambition. But ambition without a settlement layer is just a speculative trade. And as any veteran knows: trades that can’t be audited get liquidated first.
Data speaks louder than sentiment. Liquidity dries up when trust breaks. Panic sells, logic buys.