The Self-Funding AI Boom
The AI industry has created a web of circular investments. Companies like OpenAI, Nvidia, AMD and major cloud providers are investing billions in each other while simultaneously becoming customers and suppliers. This interconnected structure raises questions about whether demand is real or artificially inflated.
How the Investment Circle Works
AI companies need chips and infrastructure. Chip makers need customers. So they invest in each other—Nvidia takes stakes in its customers, who then buy Nvidia chips. OpenAI signs massive deals with suppliers, who receive equity in return. Each announcement boosts stock prices, creating a self-reinforcing loop.
The pattern appeared clearly when Nvidia owned 5% of CoreWeave, a company that buys Nvidia chips and rents them out. When CoreWeave's IPO struggled, Nvidia offered to anchor the deal—essentially funding its own customer.
Why the Deals Matter
These arrangements serve multiple purposes: securing supply chains, locking in customers, and aligning incentives. OpenAI committed to half the world's memory capacity, pledged $100 billion with Nvidia who will supply chips back, and signed AMD deals worth tens of billions—with AMD offering OpenAI the right to buy 10% of its stock for cheap.
Amazon invested $8 billion in Anthropic, which committed to using Amazon's cloud and custom chips. Google followed with its own multibillion-dollar investment and infrastructure deal. Both now have financial stakes in Anthropic's success while providing the tools it needs to operate.
The Scale Problem
McKinsey forecasts $5.2 trillion in AI infrastructure spending through 2030. The industry needs $2 trillion in annual revenue just to justify it. OpenAI has $13 billion in revenue today and burns cash rapidly. Anthropic is smaller. Nvidia is profitable, but not enough to fund everything.
The power requirements are staggering: OpenAI's Stargate project alone needs 10 gigawatts—enough to power 26 million Americans. Total OpenAI commitments require 23 gigawatts, equivalent to 23 nuclear power plants. The last U.S. nuclear reactor took over a decade to build and came online in 2024. None are currently under construction.
The GPU Market Signal
Early cracks are appearing. Nvidia B200 chip rental prices dropped from $3.20 to $2.80 per hour in months. Older A100 chips now rent for as low as 40 cents per hour—below break-even for many operators.
If demand doesn't materialize, stranded assets could pile up—data centers sitting half-empty, GPUs never earning back their cost, similar to unused fiber-optic cables from the telecom bubble.
The Monetization Challenge
OpenAI claims 700 million weekly users, but only 5% pay. Enterprise contracts drive most revenue, yet fewer than 15% of business AI pilots succeed. There haven't been mass AI-driven layoffs—only graphic designers, copywriters, and some junior coders have seen employment declines.
The question isn't whether AI works. The question is whether anyone can build a profitable business at this scale.
The Financial Reality
The big tech firms remain financially solid—projected to generate over $200 billion in free cash flow next year, even after capital expenditures. Balance sheets are strong. Valuations sit around 35x forward earnings versus 60x during the dot-com bubble.
Nvidia's largest customers are also its investment targets, using Nvidia's money to buy Nvidia's products. The OpenAI-Nvidia deal represents about 13% of Nvidia's expected 2026 revenue—significant but not overwhelming. Many deals include performance-based conditions, allowing companies to back out if monetization lags.
The Long-Term Outlook
The profits might not flow to model builders but to businesses using the models. AI could boost economy-wide productivity while the labs themselves struggle to monetize—leaving investors, lenders, and utilities holding stranded assets in a buildout that outpaced demand.