Big tech companies are spending $364 billion on AI infrastructure in 2025, but the financial model behind this boom operates like a circular money machine. Wall Street loans billions to neocloud companies who buy Nvidia chips. Which then serve as collateral for bigger loans. Capital expenditure outpaces revenue by 16 to 1, creating a black hole where money flows in but proportional returns remain uncertain.
Core Reality:
- AI infrastructure spending now drives half of US GDP growth, exceeding dot-com bubble levels
- Tech giants spent $560 billion on AI infrastructure but generated only $35 billion in revenue
- AI chips depreciate in 1-3 years but get written off over 5-6 years, hiding true replacement costs
- Neocloud providers like CoreWeave show massive backlogs that represent commitments, not actual revenue
- The market is already repricing: CoreWeave down 57%, Oracle down 34% from recent peaks
How the Circular Financing Loop Works
Wall Street loaned $11 billion to neocloud companies based on one asset: Nvidia chips.
Those companies used the loans to buy more Nvidia chips. Nvidia invested in those same companies. The chips became collateral for bigger loans.
This creates artificial demand signals. When Microsoft buys from neoclouds, those neoclouds buy from Nvidia. Nvidia’s revenue climbs, and the market interprets this as proof of sustainable AI demand.
Bottom line: The AI infrastructure boom operates as a closed loop. Where the product finances the customer who buys more product.
Why the Math Stops Working in 2026
Microsoft, Meta, Amazon, and Google will spend $364 billion on AI data centers in 2025. That spending now contributes more to US GDP growth than consumer spending.
For this to make economic sense, AI revenues need to grow from $20 billion to $2 trillion by 2030. That’s a 100-fold increase in five years.
The five biggest tech companies invested $560 billion in AI infrastructure over two years. They generated $35 billion in AI revenue combined.
Capital expenditure outpaces revenue by 16 to 1.
Key insight: The revenue growth required to justify current spending levels has no historical precedent at this scale.
What Is the Hidden Depreciation Problem?
AI chips have a useful lifespan of one to three years. Companies depreciate them over five to six years.
Nvidia now releases new chips annually instead of every two years. AMD followed. Amazon cut server lifespans from six years to five after studying the pace of technology development.
Microsoft spends $80 billion annually on AI infrastructure. If half goes to computing hardware with a true three-year lifespan. Microsoft faces $13 billion in annual replacement costs.
By depreciating over six years, reported depreciation is only $6.5 billion.
That $6.5 billion annual cushion subsidizes OpenAI’s infrastructure during the years. When customer relationships form and revenue remains theoretical.
Reality check: Accounting depreciation schedules hide the true cost of keeping AI infrastructure current. Creating a $6.5 billion annual gap between reported and actual expenses.
How Microsoft Creates Artificial Demand Signals
Microsoft signed $33 billion in deals with neocloud GPU providers. The company spends $200 million monthly on external GPU compute despite having its own datacenter teams.
The strategy: rent external GPU data centers for internal use so Microsoft rents its own facilities to other customers.
This creates artificial demand signals. When Microsoft buys from neoclouds, those neoclouds buy from Nvidia. Nvidia’s revenue climbs, and the market interprets this as proof of sustainable AI demand.
The demand is real. The question is whether this reflects end-user value or a game theory trap where no company stops spending.
The pattern: Tech giants manufacture demand signals through circular transactions. That make AI adoption look stronger than actual end-user revenue suggests.
Why Backlog Numbers Are Misleading
CoreWeave reported a 71.5% year-over-year increase in backlog. The company plans to spend over $30 billion in capex during 2026.
Backlog is not revenue. It represents future commitments that get renegotiated, delayed, or canceled if market conditions shift.
CoreWeave shares dropped 57% from their June peak. Oracle fell 34% from September highs despite aggressive data center expansion.
Market correction: Investors are repricing AI infrastructure faster than the infrastructure generates returns. Separating commitments from actual cash flow.
How AI Spending Compares to Historical Infrastructure Booms
Current AI spending exceeds the internet boom’s peak relative to GDP. When adjusted for the shorter useful life of AI chips versus physical infrastructure. AI spending surpasses even the railroad buildout of the 1860s-1870s.
AI capital spending accounts for half of US GDP growth.
Railroad companies suffered through multiple panics and bankruptcies before stabilizing decades later. Telecom stocks crashed 92% after the dot-com bust and remain 60% below their peak 25 years later.
The builders captured almost none of the economic value they created. The value went to the companies that used the infrastructure after prices collapsed.
Historical pattern: Infrastructure builders rarely capture the value they create. Returns flow to companies that use the infrastructure after prices collapse and capacity stabilizes.
What the Magnificent 7’s Spending Concentration Means
Six years ago, the Magnificent 7 represented 10% of total S&P 500 capital expenditure. Today they represent 30%.
AI-related stocks account for:
- 75% of S&P 500 returns
- 80% of earnings growth
- 90% of capital spending growth since ChatGPT launched in November 2022
Your index fund is concentrated in companies locked in a spending arms race. With no clear path to proportional returns.
This is a game theory problem. The optimal strategy would be moderate, coordinated investment. Each company fears being left behind. Forcing all players into aggressive spending that destroys the collective profit pool.
Portfolio risk: Index investors hold concentrated exposure to companies. Trapped in competitive spending dynamics that threaten long-term profitability.
What This Means for Your Next Twelve Months
The AI infrastructure boom is not a bubble in the traditional sense. Bubbles pop when speculation exceeds fundamentals.
This is a black hole. Capital flows in, financial statements show growth. But the math requires exponential revenue increases that never materialize at the required scale.
The infrastructure will eventually create value. The question is whether current investors capture any of it or whether they’re funding the foundation for someone else’s profit in 2030.
The repricing has started. CoreWeave down 57%. Oracle down 34%. The market is learning to distinguish between revenue and backlog. Between depreciation schedules and actual replacement costs, between demand signals and circular financing.
If you’re allocating capital in technology-dependent markets. The next twelve months will separate those who understood the difference between infrastructure investment and infrastructure returns.
Strategic implication: Current AI infrastructure investors face the same fate as telecom builders in 2000. They’re funding valuable infrastructure. But capturing minimal returns because competitive dynamics destroy profit pools before revenue scales.
Frequently Asked Questions
Is the AI infrastructure boom a bubble?
No. This is worse than a traditional bubble. Bubbles pop when speculation exceeds fundamentals. The AI infrastructure boom operates as a black hole where capital flows in.
Financial statements show growth, but the math requires 100-fold revenue increases with no historical precedent. Current spending exceeds dot-com bubble levels relative to GDP.
How does the circular financing loop work?
Wall Street loans billions to neocloud companies using Nvidia chips as collateral. Those companies use the loans to buy more Nvidia chips.
Nvidia invests in those same companies. When Microsoft buys from neoclouds, those neoclouds buy from Nvidia. Creating artificial demand signals that the market interprets as proof of sustainable AI demand.
Why do AI chips depreciate faster than companies report?
AI chips have a useful lifespan of 1-3 years because Nvidia releases new chips annually. Companies depreciate them over 5-6 years.
This creates a gap between reported depreciation and actual replacement costs. For Microsoft, this gap is $6.5 billion annually, subsidizing infrastructure during years when revenue remains theoretical.
What does CoreWeave’s backlog mean?
Backlog represents future commitments, not revenue. These commitments get renegotiated, delayed, or canceled if market conditions shift.
CoreWeave reported a 71.5% year-over-year increase in backlog but shares dropped 57% from their June peak because investors are learning to distinguish between commitments and actual cash flow.
How does AI spending compare to historical infrastructure booms?
Current AI spending exceeds the internet boom’s peak relative to GDP and surpasses even the railroad buildout of the 1860s-1870s when adjusted for shorter useful life.
AI capital spending accounts for half of US GDP growth. Railroad companies suffered multiple panics before stabilizing. Telecom stocks crashed 92% after dot-com and remain 60% below peak 25 years later.
Why are tech companies trapped in a spending arms race?
This is a game theory problem. The optimal strategy would be moderate, coordinated investment. Each company fears being left behind.
Forcing all players into aggressive spending that destroys the collective profit pool. The Magnificent 7 now represent 30% of S&P 500 capital expenditure, up from 10% six years ago.
What happens to infrastructure builders in these booms?
Infrastructure builders captured almost none of the economic value they created in previous booms. The value went to companies that used the infrastructure after prices collapsed.
Railroad companies suffered bankruptcies. Telecom stocks remain 60% below peak 25 years after dot-com. Returns flow to infrastructure users, not builders.
Should I adjust my portfolio based on this analysis?
Your index fund is concentrated in companies locked in a spending arms race. AI-related stocks account for 75% of S&P 500 returns, 80% of earnings growth, and 90% of capital spending growth since ChatGPT launched.
The market is repricing: CoreWeave down 57%, Oracle down 34%. The next twelve months will separate infrastructure investment from infrastructure returns.
Key Takeaways
- AI infrastructure spending operates as a circular money machine. Where Wall Street loans fund Nvidia chip purchases that serve as collateral for bigger loans, creating artificial demand signals
- Capital expenditure outpaces revenue by 16 to 1. Requiring a 100-fold revenue increase to $2 trillion by 2030 with no historical precedent at this scale
- AI chips depreciate in 1-3 years but get written off over 5-6 years. Creating a $6.5 billion annual gap between Microsoft’s reported and actual replacement costs
- The market is repricing AI infrastructure faster than returns materialize. With CoreWeave down 57% and Oracle down 34% from recent peaks
- AI spending exceeds dot-com bubble levels and accounts for half of US GDP growth. Mirroring railroad and telecom buildouts where infrastructure builders captured minimal value
- Index investors hold concentrated exposure to the Magnificent 7, which now represent 30% of S&P 500 capital expenditure and are trapped in competitive spending dynamics
- Current AI infrastructure investors face the telecom builder fate. Funding valuable infrastructure while competitive dynamics destroy profit pools before revenue scales to justify spending levels
