AI Investment Debt Risks: Hidden Liabilities and Systemic Threats

AI is currently a critical area for investors to understand, particularly concerning hidden debt obligations and their potential impact on financial stability. Ed Zitron, an AI bear and researcher, highlights significant concerns about the financial underpinnings of major AI players.

The AI Investment Landscape: A Narrow Foundation

Investors are pouring money into companies like Microsoft, Google, and Amazon, believing it supports diverse AI demand. However, Zitron argues that a substantial portion of this capital expenditure (capex) is actually propping up two specific, currently unprofitable, and unsustainable companies: OpenAI and Anthropic.

Key financial observations include:

• Google Cloud Revenue: UBS estimates that 27% of Google Cloud's revenue this year will come from OpenAI and Anthropic, projected to rise to over 48% next year, totaling over $124 billion.
• AWS Revenue: Barclays indicates that 13% of AWS revenue this year will be from OpenAI and Anthropic, increasing to 18% next year.
• Microsoft Intelligent Cloud Growth: In 2025, 69% of the year-over-year growth in Microsoft's intelligent cloud segment was attributed to OpenAI. Without this, growth would have been a mere 8%, barely above inflation.

This concentration of revenue from a few, financially precarious entities raises questions about the true health of these tech giants' AI divisions.

The "Scandal" of Circular Financing

The practice of major tech companies investing in AI startups, knowing those startups will then use the investment to purchase their products or cloud services, is a form of circular financing. While not illegal, it is extraordinarily risky. This mirrors a "seller financing" model, where the seller provides the loan for the buyer to purchase their product.

Examples of this include:

• Nvidia's investments in companies that then buy Nvidia chips.
• Google selling TPUs (Tensor Processing Units) to Anthropic, then renting them back to Anthropic through Google Cloud, effectively doubling up on revenue.

This creates a situation where the demand for AI infrastructure is artificially inflated by the very companies providing it. While these financial arrangements are disclosed, often buried deep in financial reports, the question shifts from legality to the systemic risk they create.

The Looming Debt Crisis in AI

The massive capex required for AI development means even giants like Google are experiencing negative cash flow for the first time since going public. This necessitates reliance on banks for funding. However, the speculative nature of AI makes direct bank lending risky due to regulatory scrutiny.

To circumvent this, banks are pushing this risk into the broader financial system through three main mechanisms:

1. Wholesale Funding: Banks lend to "shadow banks" (private banks not subject to the same regulations), which then lend to AI companies. This provides plausible deniability for regulated banks. Pension funds are a significant source of capital for these shadow banks.
2. Credit Risk Transfer: Banks make loans to corporate borrowers but use financial engineering to package and sell the risk off their books. This is reminiscent of mortgage-backed securities in 2008, with pension funds often being the buyers of this packaged debt.
3. Originate to Distribute: Banks act as middlemen, bundling corporate AI infrastructure loans into complex financial products called Collateralized Loan Obligations (CLOs). They then sell pieces of these CLOs to institutions like life insurance companies and pension funds, pocketing fees while offloading risk.

This "originate to distribute" model often involves mezzanine debt, a middle layer of risk between equity and senior debt. Investors in mezzanine debt accept higher risk than senior debt holders for potentially greater returns, but are still protected by the equity layer. This allows banks to appear less exposed on paper, even though the systemic risk remains.

Accounting Practices and Hidden Liabilities

Ed Zitron points to potentially misleading accounting practices that obscure the true financial state of AI companies:

• EBITDA (Earnings Before Interest, Taxes, Depreciation, and Amortization): This metric allows companies to present a picture of profitability detached from real-world costs like equipment replacement. The accusation is that AI companies are not being honest about the lifespan of their chips, claiming five to six years when they may need replacement in three, thus pushing profitability further out. Warren Buffett and Charlie Munger have criticized EBITDA, with Munger calling it "bullshit" for its ability to hide massive upfront costs.
• Adjusted Earnings/EBITDA: This involves further carving out expenses, such as stock-based compensation for employees, to make financial performance appear better. This practice, while not illegal, can obscure actual cash expenses.

The core issue is whether revenue will materialize before the debt becomes due. If chips need more frequent replacement than stated, the capital expenditure requirements increase, pushing profitability further into the future.

The "Too Big to Fail" Dilemma and Systemic Risk

The sheer scale of investment in AI, coupled with the interconnectedness of major tech companies, raises concerns about a "too big to fail" scenario. If OpenAI and Anthropic were to fail, it would significantly impact Google, Microsoft, and potentially the entire internet infrastructure.

The current situation is described as even more terrifying than the 2008 financial crisis because:

• AI is seen as a national security issue: Companies are lobbying governments to view AI as critical infrastructure, implying that they cannot be allowed to fail, potentially leading to government bailouts.
• Global Instability: This financial fragility coincides with other global instabilities, such as the yen's volatility and the US Treasury's efforts to stabilize its bond market.
• Public Sentiment: There is growing anti-AI sentiment, fueled by the perception of financial shenanigans and the widening gap between the wealthy AI elite and the struggling average person. If the AI bubble bursts and causes economic damage, public outrage could be severe.

The Timeline and Investment Strategy

The historical pattern of technological revolutions suggests a significant gap between initial investment and sustained profitability. While AI is a real and transformative technology, the revenue generation may take much longer than the debt can sustain.

Investors are urged to consider:

• The "Inheritance Generation": The truly successful companies often emerge in the second wave, building on the infrastructure laid by the first, often debt-laden, generation.
• Diversification: Given the uncertainty and the difficulty of timing the market, a diversified investment strategy is crucial to mitigate risk.
• Risk Profile: Individuals should assess their own exposure to this systemic risk, particularly through 401ks and pension funds, and consider how to protect themselves if a downturn occurs.

The current AI boom is characterized by immense capital raising, risk diversification into the broadest possible markets, and financial reporting that appears clean on paper but hides significant underlying risk. The scale of this private credit market is difficult to ascertain, but signs of distress are already appearing. The world is making a massive bet on AI, but the question remains whether the revenues will arrive fast enough to justify the unprecedented debt.