AI infrastructure debt crisis, miner leverage, and the disappearance of "liquidation liquidity"

Author: Anita; Source: X, @anitahityou

If you only look at tech news in 2025, the world seems bright: AI investment continues, North American data center construction accelerates, and crypto miners have finally "emerged from the cycle," successfully transforming their previously highly volatile mining business into stable AI computing power services.

But in Wall Street's credit departments, the atmosphere is completely different.

Credit investors aren't discussing model performance or caring which generation of GPUs is more powerful. They stare at the core assumptions on their Excel spreadsheets and begin to feel a chill: We seem to be using a 10-year mortgage financing model to buy a fresh product with a shelf life of only 18 months.

A series of reports by Reuters and Bloomberg in December revealed just the tip of the iceberg: AI infrastructure is rapidly becoming a "debt-intensive industry." But this is only the surface; the real crisis lies in a deep-seated **financial structural mismatch**—when highly depreciated computing power assets, highly volatile miner collateral, and rigid infrastructure debt are forcibly tied together, a hidden chain of default transmission has already formed. I. Deflation on the Asset Side: The Cruel Revenge of "Moore's Law" The core logic of debt financing is the Distributed Cash Flow Coverage Ratio (DSCR). For the past 18 months, the market assumed that AI computing power rentals would be as stable as rent, or even as inflation-resistant as oil. Data is ruthlessly shattering this assumption. According to data tracked by SemiAnalysis and Epoch AI in Q4 2025, the cost of AI inference per unit has decreased by 20–40% year-over-year over the past year. The widespread adoption of model quantization and distillation techniques, along with improved efficiency of application-specific integrated circuits (ASICs), has led to an exponential increase in the efficiency of computing power supply. This means that the so-called "computing rent" has a natural deflationary nature. This constitutes the first **duration mismatch**: the issuer purchases GPUs at the 2024 peak price (CapEx), but locks in a rental yield curve destined to plummet after 2025. If you are an equity investor, this is called technological progress; if you are a creditor, this is called collateral devaluation. II. The Distortion of Financing: Packaging Venture Capital Risks as Infrastructure Returns If returns on assets are thinning, a rational approach to liabilities should be more conservative. However, the reality is quite the opposite. According to the latest statistics from The Economic Times and Reuters, total debt financing for AI data centers and related infrastructure is expected to surge by 112% in 2025, reaching $25 billion. The main drivers of this surge are Neo-Cloud vendors such as CoreWeave and Crusoe, as well as mining companies undergoing transformation, which are extensively adopting Asset-Backed Lending (ABL) and Project Finance. This fundamental shift in financing structure is extremely dangerous: In the past, AI was a game for tech VCs; failure meant losing all equity. Now, AI has become a game for infrastructure; failure means debt default. The market is mistakenly placing high-risk, high-depreciation technology assets (Venture-grade Assets) into a low-risk financing model (Utility-grade Leverage) that should belong to highways and hydroelectric power plants. The most vulnerable link lies with crypto miners. The media likes to praise miners' transformation to AI as "risk mitigation," but from a balance sheet perspective, it's a **risk accumulation**. A review of data from VanEck and TheMinerMag reveals a counterintuitive fact: the net debt ratio of leading listed mining companies in 2025 has not substantially decreased compared to the 2021 peak. In fact, the debt of some aggressive mining companies has surged by 500%.

How did they do it?

1. Left hand (asset side): Still holding highly volatile BTC/ETH, or using future hashrate revenue as implicit collateral. Right side (liabilities): Issuing convertible notes or high-yield bonds to borrow US dollars to purchase H100/H200. This is not deleveraging, it's a rollover. This means miners are playing a "double leverage" game: using cryptocurrency volatility as collateral to gamble on GPU cash flow. In favorable conditions, this is double the profit, but once the macro environment tightens, both "crypto price decline" and "decline in hashrate rental" will occur simultaneously. In credit models, this is called "correlation convergence," a nightmare for all structured products. IV. The Missing Repo Market What wakes credit managers up in the middle of the night isn't the default itself, but the subsequent liquidation. In the subprime mortgage crisis, banks could at least auction off foreclosed properties. But in AI computing power financing, if a miner defaults, and creditors repossess 10,000 H100 graphics cards, who can they sell them to? This is a secondary market with severely overestimated liquidity: **Physical Dependence:** High-end GPUs aren't simply plug-and-play; they heavily rely on specific liquid-cooling racks and power densities (30-50kW/rack). **Hardware Obsolescence:** With the release of NVIDIA Blackwell and even Rubin architectures, older cards face **non-linear depreciation**. **Buying Vacuum:** During systemic sell-offs, there are no "lenders of last resort" willing to buy outdated electronic waste. We must be wary of this "collateral illusion"—the LTV on paper may seem safe, but the secondary repo market capable of absorbing billions of dollars in selling pressure simply doesn't exist in reality. This isn't just an AI bubble; it's a failure of credit pricing. To clarify, this article doesn't deny the technological prospects of AI, nor does it deny the real demand for computing power. What we're questioning is a flawed financial structure. When deflationary assets driven by Moore's Law (GPUs) are priced like inflation-hedging real estate; when miners who haven't truly deleveraged are financed as high-quality infrastructure operators—the market is actually conducting a credit experiment that hasn't been fully priced in. Historical experience has repeatedly shown that credit cycles tend to peak earlier than technology cycles. For macro strategists and credit traders, the primary task before 2026 may not be predicting which large model will win, but rather re-examining the true credit spreads of "AI Infra + Crypto Miners" combinations.