Hook
A local government releases a roadmap with aggressive penetration targets—>70% by 2027, >90% by 2030—and a 2600 billion yuan industry goal. Yet the document contains zero code, zero contract logic, zero technical specification beyond buzzwords like “next-generation intelligent terminals and agents.”
I’ve seen this pattern before. It’s the same feeling I had in 2017 when the 0x protocol whitepaper promised decentralized exchange utopia without addressing reentrancy in the relayer node.
Code doesn’t care about your feelings. A whitepaper without audit trails is a marketing deck. A policy without technical depth is a political statement. Chengdu’s “AI+” Action Plan reads exactly like a typical VC-backed DeFi project’s litepaper: high TVL dreams, zero proof-of-reserve.
Context
Chengdu, a major city in western China, has unveiled an ambitious artificial intelligence plan. Its centerpiece: by 2027, “next-generation intelligent terminals and agents” will achieve over 70% penetration in key industries; by 2030, over 90%. The target industry scale is 2600 billion yuan (about $360 billion), implying a compound annual growth rate exceeding 30%.
The plan relies on “scenario-driven + policy subsidy” models, with a “Dual 100” program (100 innovative products and 100 demonstration scenarios) and 20 annual benchmark scenarios. It aims to cultivate over 700 AI enterprises.
But here’s where my DeFi yield strategy training kicks in. I don’t evaluate a liquidity pool by its marketing page; I check the actual smart contract, the impermanent loss math, the withdrawal mechanisms. Applying the same rigor to this policy reveals a structure that looks promising on the surface but lacks the code-level verification that separates real value from rhetorical capital.
Core: A Seven-Dimension Audit
I treated this policy as I would a new AMM protocol’s documentation. I broke it down into seven technical dimensions that map directly to blockchain and DeFi engineering concepts. Each dimension gets a confidence rating based on information available, similar to how I rate a smart contract’s security before allocating capital.
Dimension 1: Tech Stack Analysis (Smart Contract Architecture)
Conclusion: The plan defines no actual model architecture, no training framework, no chip design. It’s like a protocol that says “we use smart contracts” but doesn’t specify Solidity vs. Vyper, which VM, or how to handle oracle price feeds.
Evidence: The document uses only broad terms: “next-generation AI terminals and agents.” No mention of MoE, SSM, or even basic transformer variants. No reference to inference quantization or edge compute optimization.
Hidden Info: Chengdu likely plans to use existing mature models from Huawei MindSpore, Zhipu GLM, or Baidu ERNIE. This is equivalent to building a DeFi dApp on top of Uniswap V3 codebase without forking—smart for speed, but miss out on competitive differentiation.
Unanswered Questions: What is the core technology stack behind 2600 billion? Is it just API calls to third-party LLMs, or does it require proprietary AI chips? The plan’s silence on this is a red flag, like a yield aggregator that promises high APY without disclosing the underlying protocols.
Confidence: B- (Moderate-High). Based on typical local government planning patterns, but lack of technical detail prevents accurate feasibility assessment.
Dimension 2: Commercialization Analysis (Tokenomics)
Conclusion: The monetization path is entirely “scenario-driven + subsidy”—government procurement initially, with no clear exit mechanism or market-pricing principles. This is like a DeFi protocol with 100% liquidity mining rewards and no organic fee generation.
Evidence: “Dual 100” projects and 20 annual benchmark scenarios rely on government contracts. No mention of how businesses will transition to self-sustaining revenue. The penetration target suggests consumer AI hardware (AI phones, AI PCs) which could have market pull, but enterprise/industry scenarios depend on sustained fiscal spending.
Hidden Info: Chengdu might establish AI industry funds and provide low-cost computing vouchers to reduce commercial burden. But the article fails to disclose the scale—equivalent to a project hiding its treasury multisig wallet.
Unanswered Questions: What is the ratio of government subsidy to self-generated revenue? If subsidies stop, does the industry collapse? The answer is unknown, akin to a DeFi protocol where the team controls minting keys.
Confidence: C (Moderate). Policy path exists but lacks market validation metrics.
Dimension 3: Industry Impact Analysis (Network Effect)
Conclusion: The plan will significantly boost AI adoption in western China, benefiting electronics, manufacturing, finance, and tourism. This is like a Layer 2 chain that attracts ecosystem projects through grants and marketing.
Evidence: Chengdu has a trillion-yuan electronics industry, auto manufacturing (FAW, Geely), and digital entertainment. The penetration target means consumer electronics, smart home, and wearables will directly benefit. The annual 20 benchmark scenarios will generate hundreds of billions in demand orders across system integration, data labeling, and AI consulting.
Hidden Info: Chengdu wants to nurture local champions (like Chengdu Zhiyuanhui, Chengdu Yingboge) into “header geese.” First-mover advantage likely goes to finance (Chengdu Bank) and government scenarios where the city can control deployment.
Unanswered Questions: Will the policy create barriers for non-Chengdu AI companies? How much local protectionism? Hard to measure without on-chain data on company domicile.
Confidence: A (High). Solid inference based on Chengdu’s industrial base and policy diffusion patterns.
Dimension 4: Competitive Landscape Analysis (Market Share)
Conclusion: Chengdu positions itself as “First City of AI Applications,” differentiating from Beijing (basic research), Shenzhen (hardware innovation), and Hangzhou (e-commerce cloud). But it faces direct competition from Xi’an (western computing hub) and Chongqing (smart mobility).
Evidence: Chengdu has western largest software park (Tianfu Software Park), top universities (Sichuan University, UESTC), and lower labor costs. These are comparative advantages. But Xi’an is approved as a National AI Innovation Development Pilot Zone, and Chongqing’s auto industry is catching up. Chengdu’s first-mover window is about two years.
Hidden Info: Chengdu may use “Agent” as a differentiating track because agents require scenario closure and multimodal interaction, matching Chengdu’s industrial service needs. The 2600 billion target likely includes inflation of existing electronics industry AI-enablement stats, risking statistical exaggeration.
Unanswered Questions: What is the net inflow rate of AI talent to Chengdu vs. competitor cities? Are any major AI companies (Baidu, Alibaba) planning second headquarters there? Without this data, it’s like evaluating a DEX’s TVL without knowing if top market makers have deployed.
Confidence: B (Moderate-High). Macro data supports analysis, but micro moves (company relocations) are missing.
Dimension 5: Ethics and Security Analysis (Smart Contract Auditing)
Conclusion: The policy is entirely absent of AI ethics, safety, or regulatory frameworks. This is the biggest red flag in the entire document.
Evidence: No keywords like “AI safety,” “ethical review,” “algorithm filing,” “data privacy.” While it claims to empower “a thousand industries,” it provides no guidance on access control for high-risk domains (healthcare, finance), algorithmic bias checks, or liability assignment. China’s Interim Measures for Generative AI Service Management (August 2023) requires content security review, but the policy doesn’t address how local companies will comply.
Hidden Info: The local government may rely on national regulatory coverage, creating a vacuum. The 70% penetration in terminals like smart locks and cameras implies massive personal data collection risks, but no data ethics design is mentioned.
Unanswered Questions: Do the benchmark scenarios require independent safety assessments? Who sets the standards? If an AI system causes damage (autonomous driving accident, financial misjudgment), is the enterprise or government liable? The silence speaks volumes—like a DeFi contract with no emergency pause mechanism.
Confidence: D (Moderate-Low). Inferred from omission, but typical for “industry-heavy, safety-light” policies.
Dimension 6: Investment and Valuation Analysis (Token Price Discovery)
Conclusion: The policy will have short-term catalytic effects on Chengdu-based AI concept stocks (e.g., Jiafa Education, Creative Information) in the secondary market, but long-term depends on fiscal sustainability and project return on investment.
Evidence: The 2600 billion target implies annual growth >30%, far above national AI sector growth (~15%), which will positively boost market sentiment. The “Dual 100” projects and benchmark scenarios are expected to drive government procurement, corporate investment, and bank lending. However, historical compliance rates for local government tech plans are often below 60%.
Hidden Info: Chengdu may set up a 100-billion-level AI industry mother fund, then use sub-fund SPVs to amplify leverage—undisclosed in the article. There’s also insider trading risk: institutions may have built positions in “Chengdu AI stocks” before the policy announcement.
Unanswered Questions: How much of the 2600 billion is from existing industry upgrades vs. pure new AI revenue? What tax incentives or direct financial support does the government offer? Without these details, it’s hard to calculate risk-adjusted yield.
Confidence: C (Moderate). Data and policy path are clear, but securities market reactions and final attainment rates are highly uncertain.
Dimension 7: Infrastructure and Compute Analysis (Layer 1 Scalability)
Conclusion: Chengdu’s computing infrastructure (Tianfu Smart Computing Center, Chengdu Supercomputing Center) is the core backbone supporting the plan, but computing cost and green energy supply may become medium-term bottlenecks.
Evidence: Chengdu has the National Supercomputing Center (about 100 PFlops) and the Tianfu Smart Computing Center (planned 1000 PFlops by 2025), leading in western China. The 2600 billion scale and >70% penetration require massive inference and training compute, especially edge AI chips. Electricity costs are relatively low due to hydropower, but carbon emissions constraints limit capacity expansion.
Hidden Info: Chengdu likely partners with Huawei Ascend ecosystem (already co-building Kunpeng ecology) to obtain compliant computing power, avoiding US chip sanctions. In the future, Chengdu may require local AI companies to prioritize using local computing centers to improve utilization.
Unanswered Questions: What is the total FLOPs demand by 2027? Can local compute satisfy it? How effective are the computing vouchers (Chengdu already has such policies) in subsidizing costs and creating a gravitational effect? Without a demand model, this is like analyzing a blockchain’s TPS without knowing transaction volume growth.
Confidence: B (Moderate-High). Public data on computing centers supports analysis, but demand-side modeling is absent.
Contrarian: Retail FOMO vs. Smart Money Skepticism
Panic sells, liquidity buys. The market will likely pump Chengdu AI concept stocks in the short term. But the smart money is asking critical questions that the retail crowd ignores.
First, 2600 billion yuan is an output target, not a market forecast. It includes revenue from existing electronics products with AI features added—statistical inflation. If you strip out the base industry, the net new AI revenue might be half that, or less.
Second, the lack of ethics and security framework is not just a compliance issue—it’s a liability risk. In DeFi, if a protocol has no timelock or pause function, it gets exploited. Similarly, any major AI incident in Chengdu could trigger a regulatory crackdown that halts the entire plan.
Third, the talent cost spiral. Chengdu AI salaries are already at the top of second-tier cities. As competition for engineers increases, the cost advantage erodes. This mirrors what happened in DeFi during the 2021 bull run: projects hired overpriced devs, burned through treasury, then collapsed when yields normalized.
Yield is the bait, rug is the hook. The 30%+ growth rate sounds like an irresistible APY. But without auditable input data and clear sustainability mechanisms, it’s a promise that could break upon first contact with reality.
Takeaway
Chengdu’s AI+ plan is ambitious, but it needs a code-first verification. I’d apply the same framework I use for DeFi protocols: check the contract (policy details), monitor the treasury (fiscal sustainability), and track the user base (actual enterprise adoption).
For now, I’ll keep a small tactical allocation: short-term long on local AI stocks during the initial hype, but with a hard stop-loss at -15%. The real yield, if any, will come from the structural arbitrage between government-created demand and efficient private sector execution. That gap is where a true strategist profits.
Code doesn’t care about your feelings. And neither does a 2600 billion yuan plan without a technical whitepaper. The question is whether we are providing liquidity to a genuine opportunity or just front-running a narrative.