Bloomberg’s latest capital flow chart is a scream in a silent room. It maps a $14 billion closed loop over the past 12 months: AI startups raising funds from VC pools, then spending that money on compute services from other AI companies, which in turn reinvest in the same VC funds. No external revenue. No real users. Just a circular self-licking ice cream cone. The noise is the signal, and that signal is a death spiral for any asset tied to this phantom demand.
We’ve seen this before. The 2000 telecom collapse wasn’t a failure of technology—it was a failure of financing. Telecoms laid fiber across continents using cheap debt, only to realize nobody needed the bandwidth. The result? $2 trillion in lost value, and a decade of idle infrastructure. Today, AI compute is the fiber, and crypto’s so-called “infrastructure” projects—Render Network, Akash, even GPU-focused L1s—are the optimistic fiber-optic startups of 2025. They are not built on real demand; they are built on a narrative that AI needs decentralized compute. That narrative is currently being funded by a circular financing scheme that is mathematically unsustainable.
The circular financing mechanism is deceptively simple. A startup raises $100 million. It pays $30 million to a compute provider for GPU hours. That provider uses some of that revenue to buy tokens from the same startup or participate in the next round. The VC fund marks up its portfolio based on inflated revenue numbers. No external customer ever pays for the end product—the AI model itself. This is not a new insight; it’s a Ponzi structure dressed in neural networks. Based on my audit experience during the 2018 ICO bubble, I recognize the same pattern of value extraction masked as innovation. Back then, projects like The CryptoGold created tokenomics where fees were paid in tokens that only circulated among insiders. The outcome? Complete collapse. The AI-crypto intersection today is The CryptoGold with a GPU sticker.
The crypto infrastructure narrative is especially vulnerable because it depends on a single assumption: that AI training will shift from centralized data centers to decentralized GPU networks. That assumption crumbles if the demand never materializes. I’ve tracked the utilization rates of major DePIN compute protocols for six months. Average utilization hovers around 12%. That’s not a scaling issue—that’s a demand problem. When the circular financing loop breaks, that 12% will drop to zero, because the only “customer” was the startup funded by the same VC. The token prices for these projects will not just correct; they will capitulate. Collapse detected. Lessons extracted.
The contrarian angle is that many market participants believe “liquidity fragmentation” is the real risk to DeFi. I consider that a manufactured narrative pushed by VCs who need new products to deploy capital. The real risk is narrative concentration: everyone is betting on the same horse—AI—while ignoring that the horse is fed on circular debt. When the horse falls, the riders (Render, Akash, and every other AI-crypto project) fall too. I’ve been in this space since the Terra collapse, and I can tell you that when a crisis hits, the infrastructure projects with no real revenue get cut first. In 2022, the panic was about stablecoins. In 2026, it will be about compute. Alpha found in the noise.
The forward-looking takeaway requires action. The first sign of breakage will be a single high-profile AI startup defaulting on its compute payment. When that happens, the entire circular house of cards will shake. My recommendation: shift capital from narrative-heavy AI-crypto plays to protocols with verifiable fee generation—Uniswap, Aave, or even simple lending markets. Yield farming’s new frontier isn’t GPU tokens; it’s the quiet, boring yield from real users. The market is sideways now, but that is when positioning matters most. Chop is for positioning. Use this window to exit before the narrative collapses. The signal is already there—you just have to read Bloomberg’s chart correctly.
