Sarah Breeden, the Bank of England’s Deputy Governor for Financial Stability, did not mince words last week. She called for an “urgent regulatory review” of the debt piling into AI infrastructure. Her reasoning? Repayment paths are “unclear.” The loans backing data centers, GPU clusters, and power grids are ballooning, yet the revenue models remain speculative. This is not a warning about technology; it is a warning about leverage. And if you have been watching crypto’s cycle of exuberance and collapse, the pattern is unmistakable.
2017’s dream is today’s regulation. Back then, ICOs raised billions on whitepapers touting “blockchain-enabled logistics” with no product, no revenue, and no code. Regulators arrived after the crash. Now, the same dynamic is unfolding in AI infrastructure, but with a twist: the debt is being underwritten by institutional banks, not retail investors. The scale is larger, the systemic risk is higher, and the regulatory response is still a step behind.
Context: The Debt Boom Nobody Is Talking About
The AI infrastructure buildout is a capital-hungry beast. A single hyperscale data center can cost $1–3 billion. Training a frontier model requires clusters of 10,000+ GPUs, each costing $30,000. Energy contracts run into decades. To finance these, companies and project sponsors are turning to banks, private credit, and even pension funds. Breeden’s warning suggests the total exposure may already be material to the financial system.
But here is the critical detail: the repayment mechanisms are rarely tied to signed contracts. Most projects assume that future compute demand will generate enough cash flow to service the debt. That is an act of faith, not a financial plan. During the DeFi liquidity crisis of 2020, I mapped cascade failure vectors across Compound, Aave, and dYdX. The same pattern emerges here: when everyone assumes someone else will provide the liquidity, no one does.
The macro backdrop amplifies the risk. Low interest rates over the past decade pushed investors into yield-chasing behavior. AI infrastructure loans, often structured as project finance with floating rates, became a natural destination. Now, with rates higher and the yield curve inverted, refinancing risk looms. The “unclear” repayment paths Breeden cited are not a footnote; they are the entire story.
Core: A Forensic Audit of the AI Debt Stack
Let me dissect this with the same forensic code skepticism I apply to blockchain projects. The first question: what is the asset class? AI infrastructure debt sits somewhere between real estate project finance and technology venture debt, but with none of the safeguards of either. Real estate has appraisals, rent rolls, and covenants. Venture debt has warrants and milestones. AI infrastructure debt often has nothing but a promise that someone will pay for compute.
I have seen this before. When I was a sophomore during DeFi Summer 2020, I watched yield farms promise 1,000% APY with no underlying revenue. The smart contracts were unaudited; the collateral was a governance token printed out of thin air. The crash was inevitable. Here, the “smart contract” is the loan agreement, and the “collateral” is a data center that may have no alternative use. If the AI bubble deflates, who buys a half-built GPU cluster?
The leverage ratios are disturbing. Private credit funds are lending at 60–70% loan-to-cost for AI data centers, often with interest-only periods. That is fine if the project comes online on time and finds tenants. But construction delays, chip shortages, or energy price spikes can turn a viable project into a zombie. The analog in crypto is the leveraged yield farm: high returns until one leg of the scheme breaks.
Liquidity-centric risk analysis demands we quantify the potential for cascade defaults. Breeden’s warning is a signal that the Bank of England is already mapping these interconnections. Banks that lend to AI projects also hold corporate bonds of tech companies that are investing in AI. Insurance companies backstop the loans through credit default swaps. If one large AI project defaults, it could trigger margin calls across multiple counterparties.
Regulatory opportunity framing is essential here. The AI infrastructure debt boom exists in a regulatory vacuum. Bank lending standards are governed by existing frameworks, but those frameworks were designed for manufacturing plants and office buildings, not for assets whose value depends on the next breakthrough in machine learning. Breeden’s call for an “urgent” review implies that the current oversight is insufficient. This is the same void that allowed the Terra-Luna collapse: no one had rules for what constituted “reserve transparency.”
Convergence predictive modeling: AI and crypto are not distinct universes. The same venture funds that invest in AI also hold crypto. The same tech giants that build data centers also hold Bitcoin treasuries. A debt crisis in AI would drain liquidity from the entire tech ecosystem. I saw this in 2022 when 3AC’s collapse triggered a cascade through CeFi lenders. The interconnectedness is real.
To crystallize the analysis, here are the signals I am tracking, drawn from the Bank of England’s likely playbook:
- P0 – Regulatory Action: Will the FCA or PRA issue formal guidance on AI loan risk weights? If yes, credit contraction begins.
- P1 – Bank Disclosure: Major UK banks disclose AI infrastructure exposure in Q2–Q3 2024 earnings. Any figure above 5% of total loans is alarming.
- P2 – First Default: A large AI project restructures or misses a payment. This is the “second confirmation” after Breeden’s verbal warning.
- P3 – Central Bank Report: The Bank of England’s Financial Stability Report formally labels AI debt as a “major risk.”
- P4 – Cross-Border Spillover: Other central banks (Fed, ECB) comment on similar risks. If the FOMC minutes mention AI financing, expect a coordinated response.
Contrarian: The Decoupling Thesis that Markets Are Ignoring
The prevailing narrative is that AI is a once-in-a-generation productivity revolution, so any debt taken to build infrastructure is justified by future returns. This is exactly what I heard about crypto in 2017 and again in 2021. “Bitcoin is digital gold, and Ethereum is the world computer.” Both statements were partially true, but they did not prevent a 90% drawdown when leverage unwound.
The contrarian angle: AI infrastructure debt may decouple from the underlying technology’s promise. The technology could be transformative, but the debt markets operate on a shorter time horizon. If interest rates remain high or if energy costs spike, the debt service becomes unsustainable regardless of the long-term potential. This is the same dynamic that crushed many crypto lending protocols: the underlying technology survived, but the leveraged positions did not.
Another blind spot: the assumption that AI compute demand will grow linearly with model capability. But what if model efficiency improvements reduce compute needs? Or what if regulation curbs training runs? The debt is priced on extrapolation, not on base case scenarios. The market is ignoring the tail risks.
Furthermore, the debt is not evenly distributed. A few large players (Microsoft, Amazon, Google) can self-finance. The problem lies with the second-tier players: startups, special purpose vehicles, and sovereign wealth fund-backed projects that lack the pricing power of Big Tech. These are the ones that will default first, and their interconnectedness to the banking system is where the systemic risk lies.
Takeaway: Cycle Positioning in a Post-Warning World
Breeden’s warning is not a prediction of imminent collapse; it is a diagnostic. The AI infrastructure debt cycle is in the “over-levered optimism” phase, analogous to crypto in early 2018 or early 2022. The path forward depends on regulatory response and the first default event. For macro investors, this signals a need to reduce exposure to AI-levered assets, especially private credit and high-yield bonds. For crypto, it is a reminder that unbacked leverage always finds a way to manifest, whether in ICOs, DeFi, or now AI.
2017’s dream is today’s regulation. 2024’s AI debt could be tomorrow’s systemic wake-up call. I have seen this movie before; the ending is written in the leverage ratios.
