The artificial intelligence boom is being powered by an unusual economic engine: a tightly closed loop in which the same capital circulates between investors, suppliers, and customers, often without producing sustainable profits.
When Oracle Corp. announced a multi-hundred-billion-dollar computing agreement with OpenAI last September, its stock surged by nearly 40 percent in a single trading session. The resulting $330 billion increase in market value exceeded Oracle’s total revenue from the previous decade combined. Weeks later, the company acknowledged that the deal itself would likely operate at a loss.
That pattern is becoming increasingly familiar across the AI sector.
Nvidia has committed roughly $100 billion to OpenAI while simultaneously selling the advanced chips that spending helps finance. Microsoft records OpenAI’s compute usage as Azure revenue while reinvesting capital into CoreWeave, another OpenAI supplier in which Nvidia also holds a stake. In effect, the same funds move through multiple balance sheets, creating the appearance of explosive growth without clear evidence of profitability.
What emerges is not a traditional market ecosystem but a circular financing structure operating at global scale.
When the numbers stop adding up
OpenAI’s financials illustrate the imbalance. The company reportedly lost about $5 billion in 2024 despite generating $3.7 billion in revenue. Estimates suggest it spends close to $9 billion annually just to keep its models running, before accounting for salaries, facilities, or research. Investor projections indicate losses could reach $14 billion this year and climb to more than $40 billion cumulatively by 2028, even as its valuation approaches $500 billion.
The strain extends well beyond a single firm. IBM estimates enterprise computing costs will rise nearly 90 percent between 2023 and 2025, with most executives attributing the surge to generative AI. According to data from Kruze Consulting, startups now spend roughly half of their revenue on cloud infrastructure and model inference, double the share recorded just two years ago.
When compute costs exceed the value created, profitability becomes theoretical rather than operational.
A familiar historical echo
The primary revenue engine often cited as justification, the sale of AI model access through APIs, has yet to demonstrate viability at scale. OpenAI’s API business reportedly generated about $1 billion last year, still operating at a loss. That raises questions about forecasts projecting nearly $3 trillion in AI infrastructure investment by 2029.
Citi estimates AI capital expenditure will soon account for more than one percent of US GDP growth, a concentration historically associated with wartime mobilization or speculative excess. The parallel to the late-1990s fiber-optic build-out is difficult to ignore, when vast infrastructure investments raced ahead of sustainable demand.
Circular revenue, concentrated risk
The structure of AI finance reinforces the concern. Microsoft invests in OpenAI, sells it cloud capacity, backs CoreWeave, and books the activity as cloud revenue. Amazon and Google follow similar models with Anthropic. Nvidia funds AI startups that, in turn, spend billions on Nvidia hardware. OpenAI itself has reportedly taken stakes in hardware suppliers while committing to large procurement agreements.
Analysts estimate Microsoft alone accounted for nearly one-fifth of Nvidia’s revenue in the most recent quarter.
Growth driven by such interdependence differs fundamentally from growth generated by independent demand. During the dot-com era, this behavior had a name: revenue round-tripping. Companies paid one another for capacity to inflate sales figures without corresponding cash flow. While the AI sector has not crossed that line outright, the resemblance is increasingly close.
The market-wide implications
Unlike previous tech bubbles concentrated in smaller firms, today’s exposure sits at the core of global markets. Seven companies, Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla, now account for more than half of S&P 500 gains since late 2022. Their capital spending rose 40 percent last year, while the remaining 493 firms in the index managed just 3.5 percent growth.
JPMorgan estimates AI-linked equities have driven roughly 75 percent of S&P returns and nearly 90 percent of recent capital expenditure growth. Any sharp correction would ripple across pension funds, retirement accounts, and household wealth tied to broad market indices.
How bubbles are identified
Distinguishing a transformative build-out from a speculative bubble often comes down to three tests: unit economics, demand independence, and capacity resilience. What is the true cost of each AI inference relative to the value it creates? How much revenue comes from customers who are not also investors or infrastructure providers? And what happens if power prices rise or performance gains plateau?
History shows that overbuilds, from railways to telecommunications, fail at the point where utilization cannot justify capital intensity.
AI is no exception.
The technology itself is undeniably powerful, with real and lasting potential. But current valuations assume revenue growth will arrive faster than depreciation, energy costs, and infrastructure saturation. The more likely outcome may not be a dramatic collapse, but a prolonged re-pricing as markets adjust to slower, more selective returns.
Some firms will adapt and endure. Others will discover that even the most advanced models cannot outrun basic arithmetic.
From AI spend to AI return
The cycle does not have to end in contraction. If enterprises and governments succeed in converting AI investment into measurable productivity gains, by redesigning workflows, re-engineering services, and capturing real efficiency improvements, the economics can shift.
When AI becomes a source of operating profit rather than a perpetual compute expense, capital spending can transition from hype-driven expansion to sustainable return.
That transition, not the size of the models or the scale of the funding, will determine whether AI’s circular economy ultimately spins outward, or collapses inward.
