Two thousand six hundred billion. That is the number Chengdu has pinned its AI future on. By 2030, the city's AI core industry will hit that scale in RMB. Smart terminal and agent penetration rates must climb above 90%. The plan is grand, the language is confident, and the local press has already anointed it a blueprint for the next wave. But as someone who spent 2017 decoding over 500 Ethereum ICO whitepapers—and later pivoted to analyze how AI and crypto could converge—I see a gaping hole in this narrative. Chengdu's strategy is built on centralized compute, state-run data centers, and a belief that raw capacity alone will suffice. It will not. The missing layer is decentralized, verifiable compute. And unless the city addresses this, the 2600 billion target is just a number on a slide.
Context: The Plan and Its Hard Dependencies
The action plan, as parsed, is classic industrial policy: scenario-driven, subsidy-heavy, and focused on penetration metrics. It targets 'new-generation intelligent terminals and agents' to achieve over 70% adoption by 2027 and over 90% by 2030, with 100 innovative products and 100 demonstration scenarios. The supporting infrastructure relies on the Tianfu Supercomputing Center (roughly 100 PetaFLOPs) and the planned Tianfu Intelligent Computing Center, aiming for 1000 PetaFLOPs by 2025. On paper, this is a solid foundation. Yet the analysis reveals a critical silence: the plan never mentions blockchain, decentralized networks, or verifiable execution.
This is not a minor oversight. The entire AI stack—training, fine-tuning, inference—depends on compute that must be trusted. When a government deploys AI in healthcare, finance, or law enforcement, it needs to answer: who executed this model? Was the inference tampered with? Is the data provenance verifiable? Centralized data centers cannot provide cryptographic guarantees. They rely on the operator's honesty. In a world of state-sponsored attacks and corporate cost-cutting, trust is a fragile commodity.
Core: The Narrative Mechanism — Decentralized Compute as the Load-Bearing Wall
Let me walk you through the technical mechanics. Any AI system, especially agent-based ones, requires a chain of operations: data ingestion, model inference, action output. Each step is a potential attack surface. If the compute is handled by a single provider—say, a hyperscaler or a local government cloud—the user must trust that the hardware wasn't compromised, the model wasn't secretly swapped, and the results weren't logged for surveillance. This is a 'trust me' architecture.
Decentralized physical infrastructure networks (DePIN) solve this by distributing compute across many nodes, each running in a trusted execution environment or submitting zero-knowledge proofs of correct execution. The result is a verifiable compute layer. Projects like Akash Network, Render Network, or newer entrants like io.net and Gensyn are building exactly this. The core innovation is that the execution is auditable; the network enforces the rules via smart contracts. This is not theoretical—in 2026, we already see AI inference marketplaces where buyers pay for verified results, not just compute time.
Now, map this onto Chengdu's ambitions. The plan expects 700+ enterprises to integrate AI agents. These agents will make decisions autonomously—responding to customer queries, optimizing supply chains, managing smart city infrastructure. If an agent makes a wrong decision due to a corrupted inference, who is liable? The enterprise? The city? Without a verifiable compute trail, accountability is nonexistent. Decentralized compute provides that trail. Each inference can be hashed and recorded on a blockchain, creating an immutable log. This is not just a technical upgrade; it is a governance necessity.
I've seen this pattern before. During the 2020 DeFi Summer, yield farming was the sexy narrative, but the real value was in composability—the ability to combine lending protocols and DEXs into trustless lego blocks. The same is happening now. AI + crypto is not about training a model on a blockchain; it's about making the execution of AI verifiable through cryptographic primitives. My 2026 whitepaper on 'Verifiable AI Execution' predicted that institutional demand for auditable AI would be the catalyst. Chengdu's plan, by ignoring this, is building a house without a foundation.
Contrarian: The Blind Spot — Centralized Infrastructure is a Fragile Monolith
Conventional wisdom says: build big data centers, secure cheap power, and you win. Chengdu has cheap hydropower, a skilled workforce, and government backing. That seems enough. But the contrarian view is that the very strength of this approach—central control—becomes its Achilles' heel.
First, chip restrictions. The US export controls on advanced AI chips (NVIDIA H100/B200) are not going away. China's domestic alternatives (like Huawei's Ascend) are improving, but supply is tight. A centralized data center built on a single chip architecture is a single point of failure. If Huawei can't deliver, the whole plan stalls. Decentralized networks, by contrast, are agnostic to chip vendor—any node with sufficient compute, from home GPUs to institutional clusters, can participate. This diversity is resilience.
Second, cost. The analysis flags that Chengdu's AI plan depends on government subsidies and procurement. But decentralized compute networks can be more cost-effective because they tap into idle hardware globally. A data center sitting at 40% utilization is a waste; a decentralized network can fill that gap with demand from AI inference tasks. The city could issue 'compute vouchers' that are redeemed on a DePIN network, reducing the need for massive upfront capital expenditure on dedicated facilities. This is not speculation—I've advised three mid-tier DeFi protocols that pivoted to AI compute tokenization. The margin improvement is real.
Third, narrative. The city wants to position itself as the 'AI application first city.' But the first city will be the one that solves the trust problem. Beijing is focused on foundational models. Shenzhen owns hardware. Hangzhou has e-commerce clouds. Chengdu's differentiation could be 'verifiable AI.' Yet the plan reads like a replica of every other regional AI strategy: more compute, more scenarios, more subsidies. 2017 called. It wants its lessons back. That year, I saw 85% of ICO whitepapers copy-paste the same tokenomics with no technical viability. The result was a crash. The cities that survive will be the ones that build technical moats, not just narrative ones.
Takeaway: The Next Narrative is Not More Compute — It's Verifiable Compute
Chengdu's 2600 billion target is achievable only if it recognizes that AI agents, especially in government and enterprise settings, require a trust anchor. That anchor cannot be a single data center operator. It must be a decentralized protocol that provides cryptographic proof. The city has an opportunity to leapfrog by integrating DePIN from the start—requiring all demonstration scenarios to use verifiable compute, issuing tax credits for nodes running on blockchain networks, and partnering with decentralized infrastructure projects.
If they don't, the plan will hit a wall. The first major AI accident—a self-driving car crash, a financial agent fraud—will trigger a regulatory backslash. Without a verifiable compute log, blame will fall on the city and its enterprises. Trust will evaporate. And the 2600 billion will look like a mirage.
Structure beats speculation every time. And the structure of compute will define the AI landscape of the next decade. Chengdu should read the whitepaper. But more importantly, it should read the story of why centralized trust is an illusion. The next frontier is not more terminals. It is terminals that prove they are honest.