The AI Capex Death Spiral: Why Tech Giants Are Running Out of Money
Summary of the video “It's Official, The AI Bubble Just Popped (Here's Why)” by George Gammon.
Major tech companies (Google, Meta, Microsoft, Amazon) have shifted from massive profits to deeply negative free cash flow due to unsustainable AI spending. They're now borrowing hundreds of billions in bonds at rising interest rates, caught in a death spiral where each new AI model requires exponentially more compute than the last, with no viable business model to justify the spending. Without a realistic return on investment or government bailout, this could trigger an economic crisis worse than the dot-com bubble.
The Profitability Collapse
Hyperscalers Went From Profit to Negative Cash Flow
Major tech companies (Amazon, Google, Meta, Microsoft, Oracle, Facebook) had free cash flow around 250 billion dollars annually, but have now plummeted into steeply negative territory. This reversal happened despite these companies being considered the most profitable in the world, driven entirely by massive AI capital expenditure spending.
The Core Problem: No Money, Need More Money
Hyperscalers have exhausted their cash reserves and must continue massive AI spending, but lack the internal cash flow to fund it. This creates an impossible situation where they need to borrow continuously while their profitability deteriorates.
AI Spending Is Not Like Past Capital Investments
Unlike traditional CapEx that becomes productive infrastructure, AI spending may turn into pure overhead with no return on investment. The key difference is that past technology investments became cheaper over time, but AI compute costs keep rising as models demand exponentially more power for incremental improvements.
The Debt Spiral Begins
Hyperscalers Issuing Record Bond Debt
Six major tech companies have already issued around 244 billion dollars in corporate debt in a single year just to fund AI CapEx. Even Nvidia, valued at trillions and considered the poster child for AI, had to borrow 25 billion dollars in June despite being theoretically the most profitable company in history.
Bond Market Showing Stress Signals
The corporate bond market is struggling to absorb the deluge of new debt issuance. Meta's bond yield premium over Treasuries has jumped from approximately 0.9% to over 1%, a 15 basis point increase, signaling that supply is overwhelming demand and investors are demanding higher compensation for risk.
Estimated AI Capex Need: Over 10 Trillion Dollars
High-end estimates from Wall Street Journal indicate tech giants will need north of 10 trillion dollars just in the next several years to maintain their AI race. This raises the critical question: can they possibly generate more than 10 trillion dollars in profit to justify this spending?
Rising Interest Rates Create Self-Fulfilling Prophecy
As bond yields rise, rational investors wait for even higher rates before buying, knowing more debt issuance is coming. This delays capital raising, forces companies to offer higher rates, and accelerates the death spiral. Companies must issue debt regardless of market conditions because they have no alternative funding source.
The Exponential Cost Problem
AI Model Improvements Require Exponential Compute Increases
When AI models improve by 2x (like ChatGPT-4 to ChatGPT-5), the additional compute required is not 2x but roughly 10x. This means each generational improvement becomes exponentially more expensive, and the cost of AI CapEx will not follow the typical deflationary pattern of technology—it will keep rising.
The Horsepower Treadmill
Each year, customers demand faster, more capable AI models (higher horsepower). If spending is 1 trillion dollars this year to build a certain capability, spending could need to reach 50 trillion dollars in two years to meet demand for 4x better models. Renting out old compute infrastructure covers only a fraction of new costs.
Old Infrastructure Cannot Offset New Costs
While companies can rent out older compute hardware, revenue from this is minimal compared to new spending requirements. If new spending reaches 50 trillion, old hardware might generate only 500 billion in rental revenue, leaving a massive funding gap.
The Circular Economy Accounting Trick
One Company's Revenue Is Another's Deferred Cost
When Tom buys chips from Nvidia for his data center, Nvidia recognizes immediate revenue and profit. Tom capitalizes those expenses and defers them over 5-10 years. This creates an accounting mismatch where the same dollar appears as profit for one company and as a deferred expense for another, inflating reported earnings during the boom.
Dot-Com Bubble Pattern Repeating
During 1998-2000, S&P operating earnings rose 30% over two years as companies cycled capital through each other. When order books pulled in 2000-2001, earnings dropped 40% in a single year. The same circular spending pattern is now occurring with AI CapEx among hyperscalers.
Four Possible End Game Scenarios
Scenario A: AI as Loss Leader with Upsell Model
Companies like Google could use AI as a loss leader to acquire customers, similar to sales funnels where initial products are sold at a loss but recouped through higher-margin products. However, this is unlikely because the lifetime value of AI customers cannot offset the exponential CapEx growth required year after year.
Scenario B: Resource and Energy Constraints
Physical limits on energy, semiconductors, or other resources force a halt to exponential spending growth. In this case, companies remain stuck with current-generation AI capability (the 50-horsepower car) rather than achieving the next-generation improvements everyone demands.
Scenario C: AI Replaces All Workers (Unlikely)
AI becomes an agent that replaces the global workforce. While this generates a massive addressable market, it creates a fatal economic flaw: if all workers are replaced, who has income to buy products? Aggregate demand collapses, making the business model unsustainable.
Scenario D: Government Bailout (Most Likely)
Tech companies frame AI as a national security issue and secure government funding to continue spending trillions despite only generating billions in revenue. This creates a private-public partnership where companies become effectively part of the government. Companies are racing to become too big to fail before they actually fail.
Broader Economic Implications
AI CapEx Is Propping Up the Entire US Economy
The massive spending by hyperscalers on AI infrastructure is currently supporting economic growth. If this spending collapses or must be curtailed, it removes a major prop from aggregate demand and could trigger a recession or worse.
Potential Equity Dilution and Asset Price Collapse
If borrowing costs become too high, companies will be forced to issue new equity to raise capital, diluting existing shareholders and potentially lowering stock prices. Since the entire economy is propped up by asset prices, a collapse in tech stocks would create massive headwinds for aggregate demand.
AI Bubble Larger Than Dot-Com in Relative Terms
The AI CapEx buildout relative to GDP is multiples larger than the telecom and tech buildout of 1999-2000, even as a percentage of the economy. This suggests the potential bust could be significantly worse than the dot-com crash.
Notable quotes
There's no more money. But they need a lot more money. — George Gammon
High-end estimates coming in north of 10 trillion dollars over the next several years. — George Gammon (citing Wall Street Journal)
Everyone knows there's a lot more coming. In other words, they have to issue more and more debt. — Travis King, Voya Investment Management