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

Author: Anita @anitahityou

If you only look at tech news in 2025, you’ll think the world is in great shape: AI investments are still ongoing, data center construction in North America is accelerating, and crypto miners have finally “come out of the cycle,” successfully transforming their originally volatile mining business into stable AI computing services.

But on Wall Street’s credit side, the atmosphere is completely different.

Credit investors are not discussing model performance, nor do they care which generation of GPU is stronger. They are focused on the core assumptions in their Excel sheets, beginning to feel a chill: It seems we are using a 10-year real estate financing model to buy a fresh product with only an 18-month shelf life.

Reuters and Bloomberg’s continuous reports in December reveal only the tip of the iceberg: AI infrastructure is rapidly turning into a “debt-intensive industry.” But this is just the surface; the real crisis lies in deep financial structural mismatches—when high-depreciation computing assets, highly volatile miner collateral, and rigid infrastructure debt are forcibly bundled together, a hidden chain of default transmission has already formed.

1. Asset-side Deflation: The Cruel Revenge of Moore’s Law

The core logic of debt is the debt service coverage ratio (DSCR). Over the past 18 months, the market has assumed that AI compute rent would be as stable as rent payments or even inflation-resistant like oil.

Data is ruthlessly shattering this assumption.

According to tracking data from SemiAnalysis and Epoch AI in Q4 2025, the cost per AI inference has decreased by 20–40% year-over-year.

• The proliferation of quantization, distillation techniques, and efficiency improvements in dedicated inference chips (ASICs) have led to exponential growth in compute supply efficiency.
• This means that the so-called “compute rent” has inherent deflationary properties.

This constitutes the first duration mismatch: debt issuers purchase GPUs at peak prices in 2024 (CapEx), locking in a rental income curve that is destined to plummet after 2025.

If you’re an equity investor, this is called technological progress; if you’re a creditor, this is collateral depreciation.

2. The Alienation of Financing: Packaging Venture Risks into Infrastructure Returns

If asset-side returns are thinning, the rational debt side should be more conservative.

But the reality is quite the opposite.

According to the latest statistics from The Economic Times and Reuters, the total debt financing for AI data centers and related infrastructure in 2025 surged by 112%, reaching $25 billion. The main drivers of this surge are “Neo-Cloud” vendors like CoreWeave, Crusoe, and transitioning mining companies, which are adopting large-scale Asset-Backed Lending (ABL) and Project Finance.

This fundamental change in financing structure is extremely dangerous:

• Past: AI was a tech VC game; failure meant zero equity.
• Now: AI has become an infrastructure game; failure means debt default.

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

3. The Miner’s “Pseudo-Transformation” and “Real Leverage”

The most fragile link appears among crypto miners. Media often praises miners’ transition to AI as “risk reduction,” but from a balance sheet perspective, this is risk stacking.

Reviewing data from VanEck and TheMinerMag reveals an counterintuitive fact: the net debt ratio of leading publicly listed mining companies in early 2025 has not substantially decreased compared to the 2021 peak. In fact, some aggressive miners’ debt has surged by 500%.

How do they do it?

• Left hand (asset side): still holding highly volatile BTC/ETH or using future compute income as implicit collateral.
• Right hand (liability side): issuing convertible notes or high-yield bonds to borrow USD to buy H100/H200.

This is not deleveraging; it’s rollover (debt extension).

It means miners are playing a “double leverage” game: using crypto volatility as collateral to bet on GPU cash flow. During good times, this yields double profits, but once macro conditions tighten, “coin price drops” and “compute rent declines” will happen simultaneously. In credit models, this is called correlation convergence, and it’s a nightmare for all structured products.

4. The Non-Existent “Repurchase Market” (The Missing Repo Market)

What keeps credit managers awake at night isn’t the default itself, but the liquidation after default.

In the subprime mortgage crisis, banks could at least auction the houses they repossessed. But in AI compute financing, if a miner defaults, what can the creditor do with those ten thousand H100 cards?

This is a secondary market with severely overestimated liquidity:

1. Physical dependency: High-end GPUs are not just plug-and-play in personal computers; they rely heavily on specific liquid cooling cabinets and power densities (30–50kW/rack).
2. Hardware obsolescence: With the release of NVIDIA Blackwell and Rubin architectures, older cards face nonlinear depreciation.
3. Buyer vacuum: When systemic sell-offs occur, there are no “last lenders” willing to take over outdated electronic waste.

We must be wary of this “collateral illusion”—the LTV on paper looks safe, but the secondary repo market capable of absorbing billions in sell pressure simply does not exist in reality.

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

It should be clarified that this article does not deny the technological prospects of AI, nor the genuine demand for compute power. What we question is the flawed financial structure.

When Moore’s Law-driven deflationary assets (GPUs) are priced as inflation-hedging real estate; when miners who haven’t truly deleveraged are financed as high-quality infrastructure operators—what the market is actually conducting is a credit experiment that has yet to be fully priced.

Historical experience repeatedly shows: credit cycles tend to peak earlier than technological cycles. For macro strategists and credit traders, perhaps the top priority before 2026 is not predicting which large model will win, but re-examining the true credit spreads of those “AI Infra + Crypto Miners” combinations.

<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