The ledger remembers what the crowd forgets. And right now, the crowd is euphoric about AI compute—a market where NVIDIA is not just the pick-and-shovel seller, but the bank, the customer, and the guarantor of its own demand. Ed Zitron, CEO of EZ Primary Research, dropped a truth bomb on CNBC that most crypto natives missed, but it’s a mirror for our own industry’s addiction to circular capital flows. He said: NVIDIA is “lending its credit.”
Let me unpack that. NVIDIA sells GPUs to companies like CoreWeave and Lambda—cloud infrastructure providers that are essentially renting compute to AI labs. But NVIDIA also helps those companies secure funding, through procurement contracts, direct investments, and even acting as a reference for debt financing. The money flows: investors lend to CoreWeave, CoreWeave buys NVIDIA GPUs, CoreWeave resells compute to OpenAI and Anthropic, who burn cash to train models that may or may not become profitable. The loop closes when those AI companies raise more money from VCs, who then invest in CoreWeave again. Everyone is pregnant with the same baby.
Context: The Circular Dependency
This is not a conspiracy theory. It’s documented. NVIDIA has invested in CoreWeave, Lambda, and other GPU cloud providers. It also offers financing terms to customers. The result: a synthetic demand curve. NVIDIA’s data center revenue hit $18.4 billion last quarter, up 279% year-over-year. But how much of that is real organic demand versus credit-fueled speculation? Zitron’s concern is that the end users—the AI labs—are still unprofitable. OpenAI lost $5.4 billion in 2023. Anthropic is burning through hundreds of millions. They are the ultimate consumers of this compute. If they fail, the entire house of cards collapses.
Core: The Tech + Values Analysis
From a technical perspective, this is a classic example of “turtles all the way down” in a capital-intensive industry. I’ve seen this pattern before. In 2017, I audited 15 ICO whitepapers during my university days in Tokyo. One project, EtherCrowd Alpha, promised a decentralized compute marketplace. It had a brilliant technical design—but the tokenomics were built on a circular loop: investors bought tokens, which funded GPU purchases, which were rented back to the network, generating revenue that was used to buy more tokens. The whitepaper looked like a perpetual motion machine. I flagged it because the demand side was imaginary. There were no real users outside the speculators. The project collapsed when the token price dropped and the loop broke.
NVIDIA’s situation is not identical, but the structural risk is the same. The demand for AI compute is concentrated in a handful of companies that are still dependent on venture capital. If those companies fail to achieve profitability—or if the next generation of models (like GPT-5) doesn’t deliver a step-change improvement—the perceived need for massive GPU clusters will evaporate. The suppliers (CoreWeave, Lambda) will be left with enormous debt and depreciating hardware. And NVIDIA, which has acted as the credit anchor, will face a wave of defaults or write-downs.
But here’s the deeper ethical layer: The industry is building a narrative of inevitability. “AI will transform everything, so compute demand is infinite.” That narrative is a self-fulfilling prophecy only if the capital keeps flowing. And capital flows because of the story, not because of the verified demand. We build walls of code to protect hearts of flesh, but in this case, the walls are made of debt and the hearts are levered to the moon.
Based on my experience founding BlockMind Academy, I’ve seen how hype cycles create false scarcity. In 2022, during the bear market, I watched crypto projects collapse because they had built entire ecosystems on the assumption of perpetual growth. The same pattern is repeating in AI. The difference is that AI is larger and more vertically integrated, which makes the crash potentially more systemic.
Contrarian: The Pragmatism Test
One could argue that NVIDIA’s strategy is rational. It’s creating a moat by embedding itself as the financial backbone of the industry. If you control the supply of GPUs and the credit that buys them, you control the entire value chain. This is a classic platform play. And it might work—if the end demand materializes. The counterpoint is that NVIDIA is not just a supplier; it’s a market maker. It is essentially creating its own demand by financing the customers who buy its products. That is not inherently bad—Apple does it with its supply chain financing. But the difference is that Apple’s end customers (iPhone users) pay real money for a product they want. AI model training is a cost center, not a revenue generator (yet). The buyers are VC-funded startups that are essentially gambling on a future payoff. The risk is that the payoff doesn’t come, and the credit cycle reverses.
Truth is not consensus, it is verification. The consensus is that AI compute is the new oil. The verification would require seeing real, sustainable profits from the end users. Until then, the circularity is a vulnerability.
Takeaway: What Crypto Must Learn
The crypto industry has been here before. DeFi summer 2020 was a similar circularity: stablecoins backed by other stablecoins, yield farming rewards paid in tokens that were propped up by the same liquidity. When the music stopped, the leverage unwound. The lesson is that any system built on a single point of demand concentration is fragile. For crypto AI projects (like Render, Akash, or IO.net) that rely on compute from centralized providers, this is a warning. If the NVIDIA-backed cloud providers collapse, the decentralized compute networks that depend on them for baseline capacity will suffer. The future is built by those who audit the present. Let’s audit this circularity before it audits us.
Education dissolves fear; fear creates scarcity. The antidote to this bubble is transparency. We need to track the actual utilization of GPU clusters, the cash flow of AI labs, and the debt covenants of cloud providers. As an industry, we should push for on-chain verification of compute usage—not just to prove demand, but to protect ourselves from the next crash. The ledger remembers what the crowd forgets. Let’s make sure that ledger shows real demand, not just credit.