In the summer of 2021, I sat in a repurposed warehouse in Prague, listening to a former Ethereum miner explain why his GPU rigs were being repurposed for machine learning. “We built for trustless money,” he said, “but the same chips now power centralized intelligence.” That man was an early employee of CoreWeave, a company that had quietly pivoted from crypto mining to AI cloud services. Today, CoreWeave’s stock is in a prolonged decline, and the threats are mounting. The irony is sharp: a company born from the decentralized ethos of blockchain is now facing the very centralization risks it once sought to dismantle.
CoreWeave’s story begins in the crypto boom of the late 2010s. Like many GPU miners, they accumulated vast arrays of NVIDIA hardware to secure Proof-of-Work networks. When Ethereum switched to Proof-of-Stake and mining became unprofitable, they had a choice: sell the hardware or find a new use case. They chose the latter, rebranding as a cloud provider focused on high-performance computing for AI. It was a brilliant tactical move—they already owned the infrastructure, and the AI market was hungry for cheap compute. By 2023, CoreWeave had secured investments from Microsoft and NVIDIA, and was valued at over $19 billion. They positioned themselves as the “anti-AWS,” offering flexible, low-cost GPU rentals to startups and researchers. For a moment, they were the darling of the AI infrastructure space.
But the same forces that made CoreWeave successful are now eroding its foundation. The company’s business model is a leveraged bet on NVIDIA’s GPU supply chain, long-term contracts, and volatile pricing. As a decentralized protocol PM, I’ve seen similar patterns in blockchain: projects that over-leverage on a single resource (like a token) often collapse when the market turns. CoreWeave’s decline is not just a stock price correction—it’s a signal that the centralized AI cloud model is reaching its limits.
The core of the problem lies in three technical vulnerabilities. First, single-supplier dependency: CoreWeave’s entire infrastructure is built on NVIDIA’s H100 and A100 GPUs. They have no chip design capabilities, no alternative suppliers. If NVIDIA tightens export controls, raises prices, or releases a superior architecture (like the B200), CoreWeave’s existing hardware depreciates overnight. During my time auditing smart contracts for DeFi protocols, I learned that a single point of failure is the fastest path to liquidation. CoreWeave is a liquidation event waiting to happen. Second, price competition that destroys margins: To win customers, CoreWeave undercuts AWS by 30-50%. But GPU utilization rates below 60% mean they can barely cover electricity and depreciation costs. In blockchain, we call this a “race to the bottom”—validators undercut each other until security is compromised. Here, the security is financial viability. Third, customer concentration risk: Microsoft is both an investor and a customer. If Microsoft decides to build its own GPU clusters (which it is), CoreWeave loses its anchor tenant. In decentralized governance, we always warn against “whale domination.” CoreWeave’s whale is Microsoft, and whales can swim away.
The market’s reaction—a prolonged stock decline—is rational. But as an evangelist for decentralization, I see a deeper lesson. CoreWeave represents the centralized promise of AI: fast, cheap, but fragile. It’s the opposite of the blockchain ideal, where power is distributed across many nodes and no single entity controls the resource. When I started “Prague Decentralized” in 2017, we taught that resilience comes from redundancy and community governance. CoreWeave has neither. It’s a single point of compute, governed by a corporate board, not a community of users. The very flexibility they advertise—short-term contracts, no lock-in—also means no loyalty. Customers will leave the moment a cheaper provider appears.
Now for the contrarian angle: CoreWeave’s decline might actually be good for blockchain-based compute networks. The same pressures that hurt CoreWeave—GPU oversupply, falling prices, demand for lower-cost inference—create conditions for decentralized alternatives like Render Network, Akash Network, or io.net. These protocols let individuals contribute idle GPUs to a global pool, earning tokens for their participation. They are the decentralized equivalent of CoreWeave, but without the centralized risk. As CoreWeave’s stock falls, investors may look for more resilient models. During the 2022 bear market, I saw similar patterns: centralized exchanges failed, but decentralized exchanges gained traction. History may repeat with compute.
Moreover, CoreWeave’s pivot from crypto to AI was a strategic retreat from the ideals of decentralization. They abandoned the “human-centric” dream of permissionless innovation for the “node-centric” reality of serving corporate AI. This is exactly what I warned against in my Prague workshops: “Build for humans, not just nodes.” CoreWeave built for their nodes (GPU clusters) and forgot the humans—the community of developers who could have been empowered through open, shared infrastructure. Instead, they became another cog in the silicon-valley machine. Their current trouble is not just about economics; it’s about a failure of values.
What does this mean for the blockchain space? First, the narrative that “crypto is dead” because miners pivoted to AI is misguided. The GPU infrastructure built for mining can still serve decentralized AI—if we design the right protocols. Second, the CorWeave story is a cautionary tale for any startup in our space: centralization of resources is a feature that becomes a bug. In DeFi, we saw this with lending protocols that became too dependent on a single oracle. Now we see it with compute. The solution is not to abandon the market, but to embed resilience through tokenomics, community governance, and open-source collaboration. I’ve advocated for these principles in EU regulatory discussions, and they are more urgent than ever.
Finally, the takeaway for builders: the future of compute is not just cheaper, it’s more democratic. CoreWeave’s decline is not the end of affordable AI compute, but the beginning of a shift toward decentralized infrastructure. I have seen the power of community-driven networks during the 2021 NFT boom, when artists minted on low-energy chains to prove provenance. That same spirit can guide the next wave of AI compute: open, transparent, and owned by the participants. Education is the ultimate yield—if we teach developers to value resilience over speed, they will choose protocols that distribute power, not concentrate it. The question we must ask ourselves: will we build another centralized tower, or a network of nodes that no single entity can topple?
