The Subprime Crisis of Computing Power: AI Infrastructure Debt Wave, Miner Leverage, and the Disappearance of "Liquidation Liquidity"

Author: Anita @anitahityou

If you only look at the tech news in 2025, you'll think the world is all good: AI investment continues, North American data center construction accelerates, and crypto miners finally "emerge from the cycle," successfully transforming their previously highly volatile mining business into stable AI computing power services.

But the atmosphere is quite different in the credit departments of Wall Street.

Credit investors weren't discussing model performance or which generation of GPUs was more powerful. They were staring at the core assumptions on the Excel spreadsheet and a chill began to run down their spines: it seemed we were 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 Retribution of "Moore's Law"

The core logic behind bond trading is the Distributed Cash Flow Coverage Ratio (DSCR). Over the past 18 months, the market has assumed that AI computing power rental costs will be as stable as rent, or even as inflation-resistant as oil.

The data is relentlessly shattering this assumption.

According to SemiAnalysis and Epoch AI’s Q4 2025 tracking data, the cost of unit AI inference has decreased by 20–40% year-over-year over the past year.

• The widespread adoption of model quantization and distillation techniques, along with the increased 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 power rent" has a natural deflationary property.

This constitutes the first duration mismatch: the issuer purchases GPUs at the 2024 high price (CapEx), but locks in a rental yield curve that is 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. Distortion on the Financing Side: Repackaging Venture Capital Risks as Infrastructure Returns

If returns on assets are thinning, a rational approach to liabilities should be more conservative.

But 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 projected to surge by 112% to $25 billion by 2025. This surge is primarily driven by Neo-Cloud vendors such as CoreWeave and Crusoe, as well as mining companies undergoing transformation, which are heavily utilizing asset-backed lending (ABL) and project finance.

This fundamental change in financing structure is extremely dangerous:

• In the past: AI was a game for tech VCs; failure meant losing all your equity.
• Now: AI has become a game of infrastructure; failure means defaulting on debt.

The market is mistakenly placing high-risk, high-depreciation technology assets (venture-grade assets) into low-risk financing models (utility-grade leverage) that should belong to highways and hydroelectric power plants.

III. Miners' "Fake Transformation" and "Real Leverage"

The most vulnerable link lies with crypto miners. The media likes to praise miners' transition to AI as "risk mitigation," but from a balance sheet perspective, it's actually an accumulation of risks.

A review of data from VanEck and TheMinerMag reveals a counterintuitive fact: the net debt ratio of leading listed mining companies in 2025 had not substantially decreased compared to the 2021 peak. In fact, the debt of some aggressive mining companies surged by 500%.

How did they do it?

• Left hand (asset side): Still holding highly volatile BTC/ETH, or using future computing power revenue as implicit collateral.
• Right side (liabilities): Issue convertible notes or high-yield bonds to borrow US dollars to purchase H100/H200.

This is not deleveraging, it's rollover.

This means miners are playing a "double-leverage" game: using Crypto's volatility as collateral to gamble on GPU cash flow. In favorable conditions, this doubles the profit, but once the macro environment tightens, both "crypto price drops" and "decreased hashrate rentals" 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 is not the default itself, but the liquidation that follows.

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 those 10,000 H100 graphics cards, who can they sell them to?

This is a secondary market with severely overvalued liquidity:

1. Physical dependence: High-end GPUs cannot be used simply by plugging them into your own computer; they are heavily dependent on specific liquid cooling racks and power density (30-50kW/rack).
2. Hardware Obsolescence: With the release of NVIDIA Blackwell and even Rubin architectures, older cards are facing non-linear depreciation.
3. Buying vacuum: When a systemic sell-off occurs, there is no "lender of last resort" in the market willing to take over outdated electronic waste.

We must be wary of this "collateral illusion"—the LTV on paper may look safe, but the secondary repo market that can absorb billions of dollars in selling pressure simply does not exist in reality.

This is not just an AI bubble; it's a failure of credit pricing.

To clarify, this article does not deny the technological prospects of AI, nor does it deny the real need for computing power. What we are questioning is the flawed financial structure.

When deflationary assets (GPUs) driven by Moore's Law are priced as 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 big model will win, but rather re-examining the real credit spreads of those "AI Infra + Crypto Miners" combinations.

<Reference>

<1>https://epoch.ai/data-insights/llm-inference-price-trends

<2>https://epochai.substack.com/p/the-epoch-ai-brief-april-2025

<3>https://semianalysis.com/2025/

<4>https://www.reuters.com/commentary/breakingviews/shaky-data-centre-tenants-could-choke-off-ai-boom-2025-12-10/

<5>https://longbridge.com/en/news/269179463

<6>https://economictimes.indiatimes.com/topic/data-center-capacity

<7>https://www.webpronews.com/ais-debt-fueled-data-center-frenzy-risks-mounting-in-2025-boom/

<8>https://www.alpha-matica.com/post/assessing-risks-in-ai-infrastructure-finance

<9>https://www.blackstone.com/news/press/coreweave-secures-7-5-billion-debt-financing-facility-led-by-blackstone-and-magnetar/

<10>https://www.prnewswire.com/news-releases/coreweave-secures-7-5-billion-debt-financing-facility-led-by-blackstone-and-magnetar-301848093.html

<11>https://www.cnbc.com/2024/05/17/ai-startup-coreweave-raises-7point5-billion-in-debt-blackstone-leads.html

<12>https://happycoin.club/en/vaneck-za-god-dolgi-bitkoin-majnerov-vyrosli-na-500-do-127-mlrd/

<13>https://www.binance.bh/en-BH/square/post/10-23-2025-crypto-news-bitcoin-miner-debt-surges-500-as-industry-gears-up-for-hashrate-

<14>https://www.aicerts.ai/wp-content/uploads/2025/02/Publications-Certification-Impact-Report-1.pdf

<15>https://www.webpronews.com/ais-debt-fueled-data-center-frenzy-risks-mounting-in-2025-boom/

<16>https://www.alpha-matica.com/post/assessing-risks-in-ai-infrastructure-finance