We didn’t build the internet to be this centralized, but here we are watching a GPU landlord’s stock tumble. CoreWeave, once the poster child of the crypto-to-AI pivot, has seen its market value erode week after week. The narrative is simple: a former cryptocurrency miner turned AI cloud provider is now struggling under the weight of competition and thinning margins. But as someone who has audited smart contracts for a decade, I see a more dangerous pattern beneath the surface—one that echoes the leveraged bets of Terra Luna and Three Arrows Capital.
Let me rewind. CoreWeave started life as a crypto mining operation, stacking GPUs to mine Ethereum. When the merge made mining obsolete, they pivoted to renting those same cards for AI training. It was a brilliant move—take existing infrastructure and rebrand it for the hottest tech trend. Microsoft and NVIDIA poured in capital, and CoreWeave became a billion-dollar unicorn. But the cracks are showing.
Open source isn’t just code; it’s a philosophy of transparency—something CoreWeave’s opaque financials lack. The company doesn’t publicly disclose GPU utilization rates, customer churn, or the average duration of its contracts. Based on my experience analyzing DeFi protocols, I know that hidden leverage is the first sign of trouble. In 2020, I wrote a series called “The Geometry of Trust” for Curve Finance, where I used geometric invariants to explain stablecoin risk. The same geometric thinking applies here: CoreWeave’s unit economics form a triangle with NVIDIA at the apex, the customer at one corner, and the electricity provider at the other. If any of these angles shift—say NVIDIA raises chip prices or a customer terminates a contract—the entire structure collapses.
Let’s examine the core. CoreWeave’s business model is renting NVIDIA H100 GPUs at a 30–50% discount compared to AWS or Azure. To make this work, they rely on two things: near-100% utilization and long-term contracts. But utilization is the silent killer. During my audit of Augur’s oracle mechanism, I found that prediction markets often overestimated liquidity, leading to slippage. Similarly, AI startups often overestimate their need for training compute. When the hype fades, they scale back, leaving CoreWeave with idle hardware. My analysis of on-chain data from lending protocols shows that asset utilization below 60% usually triggers a liquidation cascade. For CoreWeave, that cascade is a cash burn that spirals into debt.
The contrarian angle here is that everyone thinks AI infrastructure is a safe bet. It’s not. The market is treating GPU leasing like a commodity, but it’s really a high-leverage financial product. CoreWeave took on massive debt to buy GPUs, essentially betting that demand would outstrip supply forever. That bet worked during the crypto mining boom, but in AI, the supply of compute is elastic—new players like Lambda and Together AI are entering, and even NVIDIA itself is selling directly via DGX Cloud. This is the same pattern I saw in DeFi: a protocol grows fast by subsidizing demand, but when the subsidies stop, the users leave. CoreWeave is subsidizing its low prices with debt, and that debt is now due.
Decentralization is not a tech stack; it’s a principle that CoreWeave, despite its crypto DNA, has abandoned. They are a centralized landlord for centralized AI models. The irony is thick: the same engineers who once advocated for permissionless innovation are now locked into a single vendor for their compute needs. When I launched ArtChain Academy, I taught artists that ownership of digital assets depends on decentralized storage and execution. CoreWeave’s clients own nothing—they rent time on someone else’s hardware. If CoreWeave goes under, their training runs vanish. That concentration of risk is precisely what blockchain was supposed to eliminate.
Let me ground this in my own technical experience. In 2017, I audited the first versions of Augur and Gnosis. I found logic flaws in their oracle mechanisms that could lead to incorrect settlement. CoreWeave has a similar flaw: its entire revenue model relies on a single supplier—NVIDIA. If NVIDIA faces export restrictions or decides to prioritize its own cloud service, CoreWeave’s GPU pipeline dries up. I saw this in crypto lending: protocols that depended on a single price oracle were the first to collapse. CoreWeave is a single-oracle protocol for compute.
Red Flag: CoreWeave’s customer concentration is dangerously high. Reports suggest that Microsoft, which is both an investor and a customer, accounts for a significant portion of revenue. In the world of DeFi, we call this an “insider attack vector.” If Microsoft decides to internalize its GPU needs, CoreWeave loses its anchor tenant. During the Terra collapse, I wrote a post-mortem showing how leveraged positions concentrated in a single entity amplify systemic risk. CoreWeave is that entity for the AI training market.
Art isn’t about the canvas; it’s who owns it. Compute isn’t about the GPU; it’s who controls it. The current bull market in AI is masking a fundamental truth: decentralized compute networks like those built on blockchain (e.g., Akash Network, Render Network) offer a more resilient alternative. They distribute the risk across many providers, use token incentives to balance supply and demand, and are auditable on-chain. CoreWeave, on the other hand, is a black box.
My work bridging institutional investors to crypto has taught me that the biggest opportunity lies in the flaws of the centralized system. The prolonged decline of CoreWeave is not a story of one company failing; it’s a signal that the centralized AI infrastructure model is structurally fragile. The next iteration will be decentralized—not because it’s trendy, but because it’s mathematically safer. When I debuted my newsletter “The Decentralized Mind” in 2024, I predicted that the AI compute market would eventually fragment into thousands of small nodes connected by smart contracts. CoreWeave’s decline confirms that prediction.
So what’s the takeaway? If you are an institutional investor looking at AI infrastructure, look past the hype and into the code. Demand transparency on utilization rates, supplier diversity, and contract terms. Don’t accept a closed system when you can verify the math yourself. We didn’t survive the crypto winter by trusting centralized entities—we learned to audit, verify, and decentralize. That lesson is now more relevant than ever.
The bull market for centralized AI clouds is ending. The bear market for decentralized compute is about to begin.