The city of Chengdu just published its "AI+" action plan. A 2,600-billion-yuan target. A 70% penetration rate for "next-generation intelligent terminals" by 2027. A "double hundred" project list: 100 innovative products, 100 demonstration scenarios.
Read the document once. Then read it again. The code is missing.
Every timestamp in this plan is a potential crime scene. The authors forgot to define what "next-generation" means. Is it edge-side LLMs? Embodied agents? A glorified smart speaker with a chatbot? The ambiguity is not an oversight—it's a feature. Governments love elastic definitions because they make targets appear reachable.
Context: Protocol Background
Chengdu's plan is a classic public-sector whitepaper. It promises scale, growth, and ecosystem—but omits the technical stack that makes these claims verifiable. The city has existing strengths: a trillion-yuan electronics supply chain (Intel, Foxconn), a national supercomputing center (100 petaflops), and a planned 1,000 petaflops AI compute hub. Yet the policy treats AI as a black box that plugs into every industry without specifying the interface.
This is not a technology strategy. It's a procurement roadmap dressed in blockchain jargon. The "agent" term gets thrown around without a single reference to agent frameworks, multi-modal architectures, or even training frameworks like Megatron. For a crypto-native observer, this reads like a Layer2 rollup that claims "decentralized sequencing" without revealing the sequencer’s private key.
Core: Systematic Teardown of the Plan's Structural Flaws
Missing Technical Definitions
The plan sets a penetration target for "next-generation intelligent terminals" at >70% by 2027 and >90% by 2030. But what qualifies? A smartphone with an AI assistant? A factory robot with small language models? The metric is undefined. Compare this to a DeFi protocol that promises "99.9% uptime without downtime." Sounds great until you realize they measure uptime in blocks, not seconds, and exclude Oracle failures.
From my audit experience on the 0x Protocol v2 contracts, I learned that undefined metrics hide seven critical vulnerabilities. Here, the vulnerability is goalpost-shifting. If 70% penetration is defined as "any device with an AI sticker," the target is trivial. If it's "devices running local inference models," it's a moonshot. The plan deliberately avoids clarifying this because it wants both narratives—ambition for optics and deniability for failure.
Commercialization Without Exit Strategy
The "double hundred" projects rely on government procurement and subsidies. This is a centralized funding model, not a market-driven one. No mention of pricing mechanisms, customer acquisition costs, or unit economics. It's like a token project that promises "ecosystem growth" through treasury grants but never discloses the vesting schedule or the sale price of the governance token. The plan assumes that once 20 benchmark scenarios are built per year, enterprises will spontaneously adopt.
I've seen this pattern in DeFi: protocols that over-incentivize liquidity providers with high APRs but never plan for the taper. When the subsidies stop, the TVL evaporates. Chengdu's plan has no stated taper mechanism. The 2,600-billion-yuan target likely includes "traditional product + AI feature" revenue inflation—a common statistical trick that makes every factory's upgraded sensor count as "AI output."
Security and Ethics: The Silence That Screams
The plan contains zero mentions of AI security, ethical review, algorithm registration, or data privacy. For a policy that aims to penetrate 90% of terminals across healthcare, finance, and governance, this is negligence. In crypto, we call this a reentrancy vulnerability—the attacker (bad actor) can call back into the system before the state updates. Here, the state update is compliance. The plan promises rapid deployment but leaves the security audit for later.
China's Generative AI regulations require content safety checks and model registration. How does Chengdu help its 700+ enterprises comply? The plan doesn't say. This is like a Layer2 chain that launches without a sequencer upgrade path—functional today, ungovernable tomorrow. Every log line about "empowering all industries" should trigger an alarm. Silence in the logs screams louder than alerts.
Resource Constraints: The Bottleneck They Ignored
The plan's success hinges on two resources: talent and compute. Chengdu has good universities (Sichuan University, UESTC) and a national supercomputing center. But the demand from 700+ enterprises will far exceed local capacity. The plan doesn't model the total FLOPs required for 70% terminal penetration. It's like a mining pool promising to secure 30% of Ethereum hashrate without publishing its ASIC purchase orders.
Worse, the plan doesn't address chip export controls. If the US tightens restrictions on Nvidia and AMD GPUs, Chengdu's compute expansion depends on domestic alternatives (Huawei's Ascend). But the policy doesn't mention this dependency. It assumes infinite supply of silicon, which is as foolish as assuming infinite liquidity on a frozen lending pool.
Convergence with Blockchain Patterns
This plan mirrors the classic crypto project cycle: bold headline number → vague technical description → government/VC backing → community hype → slow bleed. The only difference is that here, the "community" is the taxpayers, and the "protocol" is the city administration. The 2,600-billion-yuan target will likely be recalculated or redefined after three years, just as many DeFi assets do a "supply recalculation" after a flash loan attack.
Contrarian: What the Bulls Got Right
To be fair, the plan has one thing going for it: alignment with existing industrial strengths. Chengdu does have a massive electronics manufacturing base, and AI integration could genuinely improve factory efficiency. The "double hundred" projects may produce a few successful vertical solutions—just like how some DePin projects do deliver real-world sensor utility despite the hype. The government's ability to enforce standardization in controlled environments (e.g., city governance, public healthcare) could create stable demand.
Also, the plan avoids the "foundation model arms race" trap. By not chasing GPT-5-level research, Chengdu focuses on applied AI, which has a higher short- to mid-term ROI. This is analogous to a blockchain project that skips building its own L1 and uses an existing chain (e.g., Polygon) for faster deployment. Smart, but it means the city is building on rented land.
But these strengths don't address the fundamental architectural flaws. The plan is a monolith—it assumes top-down coordination will work, while every successful decentralized system (including real AI ecosystems like Hugging Face) evolved bottom-up. The central planning authority is the single point of failure. If the government's priorities shift, the entire "protocol" forks into oblivion.
Takeaway: The Autopsy Will Be Written in Logs
Chengdu's AI+ plan is a technological whitepaper that forgot to include the code. It defines outputs without inputs, targets without metrics, and scale without sustainability. The city will likely hit some early milestones through sheer fiscal force—just like a DeFi protocol can show high TVL for three months before the ice cracks. But the real test is the bear market of attention: when initial subsidies end and the global AI hype cycle cools, will the terminals still run?
Code does not lie; it merely waits. The lies in this plan will surface as missing benchmarks, redefined targets, and silent cost overruns. Every timestamp from now until 2030 is a potential crime scene. The question is not whether Chengdu will build the AI ecosystem—it's whether they will audit it before it implodes.
Reputation is liquid; solvency is binary. The plan's solvency depends on whether the technical details catch up to the promises. Based on the evidence, I'd short the 2030 target.