The Hong Kong Stock Exchange’s ticker 02513.HK just did something rare in a sleepy summer session: it jumped 30% on a single announcement. The company behind it, Zhipu, declared plans to acquire Zhongke Jiahe and build a 1GW computing center. Markets cheered. But for anyone who has spent years watching the intersection of compute and crypto, the celebration feels like a warning.
Here is the first problem: we do not know if this Zhipu is the same Zhipu AI that powers the GLM-4 large language model, or a shell with a similar name. The lack of clarity is a red flag that should make any Web3 investor pause. Even if it is the same company—a unicorn funded by Chinese state-backed capital—the path to a 1GW data center is a path away from the values we hold dear.
The announcement is a textbook case of centralized infrastructure propaganda dressed in growth narrative. 1GW of compute is enough to run several million GPUs simultaneously. To put it in perspective, the entire Ethereum network has historically relied on a fraction of that power for Proof-of-Work. Today, the Web3 ecosystem runs on a distributed mesh of smaller providers: Akash Network, Filecoin compute, Golem, and dozens of zk-Rollup provers scattered across continents. Decentralized computing is not a theoretical ideal; it is a working model that protects against single points of failure, censorship, and control.
Zhipu’s move is the antithesis of that model. It consolidates power into a single geological location, probably in a western Chinese province where electricity is cheap but governance is opaque. The acquisition of Zhongke Jiahe, a firm with ties to the Chinese Academy of Sciences, suggests government alignment—which might be good for compliance but terrible for the permissionless innovation that blockchain requires.
Let us be specific about the technical impact. A 1GW compute center, if built for AI inference, can generate zk-proofs at a rate that would overwhelm any decentralized prover network. If the center is used for training, it will create a closed-loop advantage: the models improve faster, the data is siloed, and the community loses access to the underlying intelligence. This is exactly the dynamic we see in centralized exchange market making, where a single firm’s algorithm sets prices for everyone. The same will happen in compute.
During my work building ChainLit in 2017, I saw how a single misleading whitepaper could drain retail wallets. Today, the weapon is not a whitepaper but a press release. The 30% price jump is based on a promise of future compute, not on any verifiable benchmark. Who guarantees that the center will be used for productive AI or blockchain work? It could be used for state surveillance, censorship filtering, or simply to mine a competing token. The community has no control over it.
This is where our Pedagogical Simplification becomes critical. Let me break down what 1GW means in terms you can feel: - It can power a small city of 200,000 homes. - The annual electricity cost at $0.05/kWh is over $400 million. - The cooling alone requires a dedicated river or extreme liquid immersion systems like cold-plate or immersion. - If it uses Nvidia H100 GPUs, even with the current export restrictions, it would require ~200,000 units. If it uses Huawei Ascend 910B, the throughput for zk-SNARKs is still unproven at scale.
The probability that this compute center is built and operated with transparent smart contracts? Near zero. The probability that it will be used to support decentralized protocols like Ethereum, Solana, or Polygon? Unlikely. Instead, it will be a moat for a private AI empire.
And here is the contrarian angle that might save it: a massive compute center could be a public utility if it is governed by a DAO. Imagine if Zhipu tokenized access to the computing power via a utility token, with on-chain slashing if the center acts maliciously. That would be a step toward decentralization. But nothing in the announcement suggests that. The market is not pricing in decentralization risk; it is pricing in China-inc narrative.
For Web3 builders, this is a wake-up call. The race for compute is real, but the solution is not to build a bigger silo. It is to build better bridges between small, verifiable compute providers. We need to invest in decentralized zk-prover marketplaces, in verifiable computing on platforms like Aleo, and in community-owned infrastructure like the ones being built by the Injective and Celestia ecosystems.
Let me embed another first-hand insight: after the FTX collapse in 2022, I saw how centralized trust can evaporate in 48 hours. The same will happen to any single compute center if the government in power changes policy, or if a natural disaster strikes. Decentralized resilience is not a luxury; it is a survival requirement.
Now, let us talk about the data availability hype. Many rollups claim they need dedicated DA layers, but 99% don’t generate enough data to require a Celestia or EigenDA. Similarly, do we really need 1GW to serve blockchain compute? Most dApps today run on lightweight nodes. The real demand is for inference at the edge, not for training large models. The Zhipu center is sized for a world where every human relies on AI, but that world is 5–10 years away. Until then, it is a massive capital asset that will depreciate like a supertanker.
In the bull market of 2024–2025, euphoria masks technical flaws. The Zhipu announcement is a perfect example: everyone celebrates the size, but no one audits the cost structure. The code does not lie, but press releases do.
So what do we do? We shift our lens from admiration of scale to obsession with verification. Every token holder should demand: 1. Are the GPUs benchmarked on an open dataset? 2. Is the compute center’s energy mix disclosed on-chain? 3. Can I rent one GPU via a smart contract without asking permission?
If the answer to any of these is no, then the 30% jump is just noise.
Here is my takeaway: Community is the only chain that cannot be broken. A 1GW center is a chain of concrete and silicon that can be shattered by a single earthquake or political decree. A community-governed mesh of 10,000 small providers is resilient because its pieces are spread across cultures, jurisdictions, and ideologies.
We must resist the temptation to bow to centralized convenience. The future of Web3 compute is not a giant fortress; it is a garden—a thousand independent flowers that bloom under the sun of permissionless protocol.
Trust is earned in the bear, spent in the bull. Today, Zhipu has spent some trust. Let us see if they earn it back with transparency. Until then, keep your compute decentralized and your values uncompromised.
