By Tony Grayson | Independent Strategic Advisor | Top 10 Data Center Influencer | Former SVP Oracle, AWS & Meta | U.S. Navy Nuclear Submarine Commander (Ret.), USS Providence (SSN-719) | Stockdale Award Recipient

Published: November 13, 2025 | Updated: February 22, 2026 | Verified: February 22, 2026

In 30 Seconds:

The Bank of England's Financial Stability Report called AI equity valuations "materially stretched." JP Morgan reports AI debt now accounts for 14% of its investment-grade index -- surpassing U.S. banks.

The structural parallels to 2008:

1. SASB binary risk -- single hyperscaler tenants with no portfolio diversification

2. SPV off-balance-sheet debt -- same mechanism as CDOs, different asset class

3. Asset-liability mismatch -- 15-year building debt on 3-year GPU technology

4. CapEx-to-revenue gap -- $800B in private credit required vs. $12B in actual AI revenue

The bottom line: The market knows what this pattern looks like. It is funding it anyway.

A Note on Perspective: 

The market consensus says AI infrastructure investment is justified by long-term demand fundamentals -- that the CapEx is rational given where AI adoption is heading. This post does not dispute the demand thesis. It disputes the financing structure. Demand being real does not make the debt instruments sound. The same was true of housing in 2005.

According to JP Morgan, debt tied to AI companies now accounts for 14% of its investment-grade index—surpassing U.S. banks as the dominant sector.

While hyperscalers drove AI infrastructure spending to nearly $200 billion in 2024, regulators have escalated their warning. The Bank of England's latest Financial Stability Report called public equity valuations "materially stretched" and explicitly flagged rising financial stability risks if the spending cycle turns. The financing boom looks eerily like the setup for a 2008-style AI market correction.

Key Concepts

AI Debt Bubble: The accumulation of off-balance-sheet debt financing rapid AI data center expansion, creating a gap of over $600 billion between infrastructure spending and actual AI revenue. Structurally analogous to the mortgage-backed securities market before 2008 -- same financial engineering, different asset class.

Shadow Debt: Off-balance-sheet financing structures that do not appear on corporate balance sheets but still create real financial obligations. In AI infrastructure, this includes SPVs, SASB loans, and CMBS offerings used by hyperscalers to fund data center expansion without impacting credit ratings.

SASB Binary Risk: The failure mode of Single-Asset-Single-Borrower loans. No portfolio diversification -- if the tenant vacates or technology becomes obsolete, the revenue stream stops and the loan fails entirely. As Dan McNamara of Polpo Capital put it: "When things go bad with SASB, they go really bad."

Asset-Liability Mismatch: The gap between GPU useful life (3-4 years) and building debt terms (15-20 years). Facilities custom-built for specific GPU generations lose collateral value far faster than debt is paid down.

The Hardware Cliff: The point at which current-generation GPU hardware is superseded, stranding billions in securities backed by facilities optimized for obsolete chip specs. Not a future risk -- already occurring as GB200s replace H200s.

CapEx-to-Revenue Gap: Only 3% of consumers pay for AI services (~$12 billion annually) while hyperscalers require $800 billion in private credit over two years. For every $1 spent on infrastructure, the market sees only a fraction returned in revenue.

Understanding the Risk on the AI Debt Bubble: Financial Engineering 2.0

To understand the financial engineering driving this boom, it helps to revisit the last crisis. The mechanism isn't new; the asset class has just changed from housing to compute.

Watch: The Warning Signs This discussion breaks down the exact risks building in the AI financing market right now.

"If there's a problem with AI data centers, like if their current chips are obsolete in five years, you could have big losses in these deals. When things go bad with SASB, they go really bad."

-- Dan McNamara, Founder, Polpo Capital

The concentration of hyperscaler debt is accelerating:

• Blackstone's Move: Recently closed a $3.46 billion CMBS offering backed by QTS data centers—larger than the entire data center CMBS market for all of 2024.

• 2025 Surge: In the first half of 2025 alone, eighteen ABS and CMBS deals totaling $13.4 billion closed.

AI Infrastructure Spending vs. Revenue: The Math Doesn't Work

The fundamental issue driving the AI bubble narrative is the gap between CapEx and revenue.

Here is the uncomfortable truth behind the valuation:

1. Revenue Reality: Only 3% of consumers currently pay for AI services, generating roughly $12 billion annually.

2. Spending Reality: Hyperscalers require $800 billion in private credit over the next two years to sustain current infrastructure spending.

As I detailed in my analysis of the AI infrastructure investment gap., the unit economics simply do not close at these debt levels without a massive shift in adoption.

We are seeing a disconnect between "valuation" and "value." As I wrote in AI Infrastructure Financing Risks., leaders must look beyond the hype cycle to understand the ground truth. Morgan Stanley is betting the industry can securitize its way to profitability, but central banks warn of "stretched valuations" in a sector where "the future is highly uncertain."

Two Paths for AI Financing: Diversification vs. The Gamble

Goldman Sachs laid out the choice clearly. Wall Street is currently betting on Path 1—using 2008-style financial engineering to fund the boom.

The Hardware Cliff: Financing Temporary Tech with Permanent Debt

If H200s and GB200s are already replacing NVIDIA's H100 chips, what happens to billions in securities backed by facilities optimized for last generation's hardware?

This is the NVIDIA Vendor Financing Trap: Hyperscalers are building permanent infrastructure for temporary technology, and financing it with off-balance sheet debt designed to hide risk.

The market has seen this movie before. In 2008, it was mortgage-backed securities. Today, the assets are real, the data centers exist, but the financial engineering is eerily familiar. The question isn't whether AI is real; it is whether the AI debt bubble can survive its own weight.

In my post on  the math on risk, I discussed the difference between calculated risk and reckless gambling. Right now, the market is gambling.

"This time, we recognize the plot... but we are still funding the sequel."

-- Tony Grayson, Top 10 Data Center Influencer | Independent Strategic Advisor

"The difference between calculated risk and reckless gambling is whether you understand the downside. Right now, the market is gambling."

-- Tony Grayson, Former SVP Oracle, AWS & Meta

Frequently Asked Questions: AI Debt Bubble and Infrastructure Financing Risks

What is the AI Debt Bubble?

The AI Debt Bubble is the accumulation of significant off-balance-sheet debt used to finance rapid data center expansion, creating potential liquidity risk. Analysts estimate a gap of over $600 billion between infrastructure spending and actual AI revenue. The core risk is a market correction where future cash flows from AI software fail to cover debt service costs incurred to build physical capacity today.

What did the Bank of England warn about AI?

The Bank of England's Financial Stability Report called public equity valuations "materially stretched" and explicitly flagged rising financial stability risks if the AI spending cycle turns. They warned the financing boom looks "eerily familiar" to the structures that preceded the 2008 financial crisis—concentrated single-asset debt, off-balance-sheet vehicles, and asset-liability mismatches.

How do Data Center SPVs (Special Purpose Vehicles) work?

A Data Center SPV (Special Purpose Vehicle) is a separate legal entity established by a parent company (like a hyperscaler) to isolate financial risk and protect corporate credit ratings. The parent transfers specific assets (like a data center facility) into the SPV, allowing them to raise capital against that project without the debt appearing on their main balance sheet—preventing downgrade of corporate borrowing power.

What is SASB financing in data centers?

SASB (Single-Asset-Single-Borrower) financing is a loan structure backed entirely by one specific property with one tenant—such as a facility leased 100% to a single hyperscaler. These loans carry higher "binary risk": if the tenant vacates or technology inside becomes obsolete before the loan matures, the revenue stream stops and the loan can fail, with no portfolio of other assets to absorb the loss.

How does hardware obsolescence impact AI infrastructure financing?

Hardware obsolescence creates an "asset-liability mismatch" where financing terms (15-20 years for buildings) exceed technology useful life (3-4 years for GPUs). If facilities are custom-built for specific generations of cooling or power density that become outdated, the underlying collateral loses value much faster than debt is paid down. NVIDIA's H200s and GB200s already replacing H100s demonstrate this risk. See also: NVIDIA Vendor Financing Infrastructure Risks.

What is the CapEx to Revenue Gap in AI?

The CapEx-to-Revenue Gap refers to the disparity between massive capital expenditures on NVIDIA chips and data center builds versus actual AI software revenue. Only 3% of consumers currently pay for AI services, generating roughly $12 billion annually, while hyperscalers require $800 billion in private credit over two years. For every $1 spent on infrastructure, the market sees only a fraction returned in revenue.

What is shadow debt in AI infrastructure?

Shadow debt refers to off-balance-sheet financing structures that don't appear on corporate balance sheets but still create financial obligations. In AI infrastructure, this includes SPVs, SASB loans, and CMBS offerings used by hyperscalers to fund data center expansion without impacting their credit ratings. This debt is "hidden" from standard financial analysis but creates real systemic risk.

How much AI-related debt is in the market?

According to JP Morgan, AI-related debt now accounts for 14% of their investment-grade index—surpassing U.S. banks as the dominant sector. Blackstone alone closed a $3.46 billion CMBS offering backed by QTS data centers. In the first half of 2025, eighteen ABS and CMBS deals totaling $13.4 billion closed. AI infrastructure spending reached nearly $200 billion in 2024.

What is CMBS in data center financing?

CMBS (Commercial Mortgage-Backed Securities) are bonds backed by commercial real estate loans. In data center financing, CMBS offerings allow investors to buy into debt secured by data center properties. Blackstone's $3.46 billion CMBS offering backed by QTS data centers was larger than the entire data center CMBS market for all of 2024—showing rapid concentration of risk.

Is AI the next 2008 financial crisis?

The Bank of England warns the financing structures look "eerily familiar" to 2008. The parallels: concentrated single-asset debt (SASB vs subprime), off-balance-sheet vehicles (SPVs vs CDOs), asset-liability mismatches (15-year debt on 3-year technology vs 30-year mortgages on depreciating homes), and stretched valuations. The mechanism isn't new—only the asset class changed from housing to compute.

What are the two paths for AI financing?

Goldman Sachs outlined two paths: Path 1 (high risk) uses single hyperscaler tenants (SASB), long-term debt-heavy leases, high AI concentration risk, and hardware lock-in. Path 2 (sustainable) uses co-location with multiple tenants, flexible shorter-term leases, diversified tech risk, and chip-agnostic architecture. Wall Street is currently betting on Path 1 -- using 2008-style financial engineering.

Who is Tony Grayson?

Tony Grayson is President & GM of Northstar Enterprise + Defense, former SVP at Oracle ($1.3B budget), AWS, and Meta (30+ data centers). He commanded nuclear submarine USS Providence (SSN-719) and received the Stockdale Award. His hyperscale infrastructure experience informs his analysis of AI financing risks and the parallels to 2008's financial engineering.

Sources:

Bank of England Financial Stability Report, December 2025 -- Bank of England

AI Boom Fueling Record Corporate Bond Sales -- Bloomberg / JP Morgan

Hedge Fund Polpo Sees Risk in AI Data Center Debt Boom -- Bloomberg / Dan McNamara

Blackstone $3.46B CMBS Data Center Refinancing -- Commercial Observer

AI's $600B Question -- Sequoia Capital

2007-2008 Financial Crisis -- Wikipedia

Commercial Mortgage-Backed Security -- Wikipedia

Special-Purpose Entity -- Wikipedia

Off-Balance-Sheet -- Wikipedia

Collateralized Debt Obligation -- Wikipedia

More from The Control Room:

AI Infrastructure Financing Risks: What JLL's 2026 Data Center Outlook Missed

Year 6 Survival Model: AI Data Center Investment Risk and Why Lease Duration Breaks the Math

Is Nvidia the Next Cisco? The 25-Year Warning Explained

Fermi Stock Crash Explained: The First Casualty of AI's Nuclear Fantasy

AI Bubble 2025: Why Herd Mentality is the Real Investment Trap

Is Your GPU Debt a Stranded Asset? Inside NVIDIA's $20B Groq Pivot

The Hidden Stack: The Real Truth About AI Infrastructure Spending

_____________

Tony Grayson is an independent strategic advisor at the intersection of AI infrastructure, nuclear energy, and national defense, and a former U.S. Navy Nuclear Submarine Commander with DOE and Naval Reactors certification.

He commanded USS Providence (SSN-719) before leading global infrastructure at scale: SVP of Physical Infrastructure at Oracle ($1.3B budget), Global Head of Design & Engineering at AWS, and Global Director of Operations at Meta (30+ hyperscale data centers). He founded and led EdgePoint Systems / Compass Quantum to a Top 10 modular data center ranking before its acquisition.

Tony is a recipient of the Vice Admiral James Bond Stockdale Award for Inspirational Leadership and serves as Veterans Chair for Infrastructure Masons, having helped transition 100+ veterans into technology careers. He is recognized as a Top 10 Data Center Influencer and serves on the advisory boards of TerraPower and Holtec International.

Read more at tonygrayson.ai