Last week, CoreWeave’s CEO stood before the press and declared that their massive AI infrastructure deployment was “on track” and that “revenue growth will ease the depreciation effect.” The crowd nodded. The stock ticked up. But I couldn’t shake the feeling that I had heard this story before. —— In 2017, the same CEO was pitching a crypto mining farm, promising that ASIC depreciation would be covered by the rising price of Ethereum. The script has changed, but the architecture of trust remains the same.
Trust the protocol, not the pitch.
CoreWeave began life as a crypto mining operation, riding the ICO wave by stacking GPUs to mine Ethereum. When the merge arrived and Proof-of-Stake eliminated mining, they pivoted — not to building a decentralized compute network, but to selling GPU time to AI startups. It was a smart business move. They repurposed their existing hardware, their cheap electricity contracts, and their data center know-how. Today, they claim to run one of the largest H100 clusters in the world, backed by Microsoft investment and a $19 billion valuation.
But let’s call this what it is: a centralized cloud built on centralized chips, with centralized pricing and zero verifiability. The crypto community should be suspicious, not celebratory.
The core of CoreWeave’s technology is not innovation in compute architecture or distributed consensus. It is simply buying NVIDIA GPUs in bulk, connecting them with InfiniBand, and renting them out at a 30-50% discount to AWS. There is no novel scheduling algorithm open-sourced, no peer-to-peer resource sharing, no cryptographic proof that your training data remains private. It is a classic rent-seeking model dressed in AI-native clothing.
Based on my audit experience in crypto mining infrastructure, I can tell you that the same fragility exists here. In 2020, I audited a high-yield farming protocol that claimed to be trustless, only to find a reentrancy vulnerability that could drain $5 million. CoreWeave’s CEO is not exposing a smart contract bug, but his financial architecture has a similar vulnerability: over-reliance on a single supplier (NVIDIA), a single demand driver (AI training), and a single narrative (scaling).
Silence is the loudest audit. When the CEO talks about “easing depreciation,” he is really saying that his capital expenses are so massive that only exponential revenue growth can prevent a balance sheet crisis. This is not a sign of health; it is a levered bet on continued AI hype.
The contrarian view that few want to hear: CoreWeave’s success may actually be bad for blockchain’s vision of decentralized compute. Why? Because it creates an illusion of “AI infrastructure” that is centralized, opaque, and non-sovereign. Every dollar flowing to CoreWeave is a dollar not flowing to Akash, Render, or other decentralized compute protocols that offer verifiable execution and permissionless access. The market is choosing the easiest path — rent a GPU from a known provider — over the harder path of building a truly open compute layer.
Code doesn’t lie, but capital does. CoreWeave’s financial statements are private. We have no way to verify that they own the number of GPUs they claim, or that their data centers have the redundancy they promise. The CEO’s words are just a pitch. In crypto, we have learned to distrust the pitch and audit the protocol. This time, there is no chain to audit.
There is also an ethical dimension. CoreWeave’s infrastructure can be used to train harmful models — deepfakes, autonomous weapons, surveillance algorithms — and the company has not published a credible Acceptable Use Policy that enforces alignment with human rights. Their only filter is business viability. In a bull market for AI, that filter is very weak.
Takeaway: The next time you hear a centralized provider promise “scale” and “revenue growth,” remember that the same words were used to sell ICOs. The decentralized compute movement must reclaim the narrative. We need verifiable, permissionless, and ethical infrastructure — not just a repackaged data center with a fresh coat of AI paint. CoreWeave’s story is not over, but the crypto community should watch it closely, not as a model to emulate, but as a cautionary tale of what happens when we prioritize speed over sovereignty.
