If there’s an AI bubble, it’s hiding in the credit market

To add a little balance to my “The New Era Isn’t a Bubble — It’s a Rerating of Reality,” it’s not that I’m naive to the risks. Every cycle hides its own fault line, and this one won’t be any different. However, I think the tail risk will persist more deeply in the credit market.

There’s a strange echo in the market these days — a low-frequency hum of capital that seems to come not from factories or consumers, but from the future itself. The AI boom, once sold as a profit flywheel of endless efficiency, has quietly become the world’s most elaborate exercise in financial time travel — where tomorrow’s expected massive cash flow is used to fund today’s infrastructure in the hope that belief alone will bridge the gap.

At first, it looked self-sustaining: hyperscalers like Microsoft, Amazon, and Google reinvesting their monstrous free cash into data centers, GPUs, and AI training clusters. But something shifted. Oracle decided to break ranks, levering itself to the hilt — a 500% debt-to-equity ratio — to join the hyperscaler Olympics. Suddenly, what was once a cash-flow-fed growth story began to smell like a credit-fed arms race. It’s as if a handful of companies are trying to build four nuclear plants’ worth of power and calling it innovation.

The irony? The “AI miracle” has generated barely enough revenue to pay for college essay subscriptions and a handful of enterprise pilots. Yet $500 billion in annual capital expenditure needs financing — now. So the question that no one on CNBC seems willing to ask is: who’s actually footing the bill for this moonshot? The answer, increasingly, is debt.

What used to be called venture capital has been reborn as venture credit. Banks, private funds, sovereigns, insurers — everyone is being enlisted to bankroll the AI gold rush. Morgan Stanley estimates nearly $3 trillion in global data center spend through 2028, half of which can’t be funded internally. That leaves a $1.5 trillion hole. Enter private credit, collateralized data centers, and “AI infrastructure bonds” — all dressed up in ESG-scented wrappers and sold to pension funds starved for yield.

We’ve seen this playbook before: telecoms in 1999, shale oil in 2013, crypto mining in 2021. Each cycle begins with a story that promises exponential growth and ends with financiers discovering exponential math doesn’t always apply to balance sheets. The new AI ecosystem — a web of interlocked promises, circular contracts, and synthetic cash flows — feels eerily similar. Oracle borrows to build data centers to host OpenAI workloads that don’t yet exist, generating revenues that investors assume will. It’s a hall of mirrors built on future belief — and belief, like credit, is only as strong as its collateral.

For the first time in modern market history, AI-related corporates now represent 14% of the U.S. investment-grade bond index, surpassing the banking sector. That’s an extraordinary shift in the financial architecture of capitalism: tech has effectively become the new financial system. But here’s the paradox — these are not banks. They’re not capitalized, regulated, or stress-tested like banks. They’re industrial conglomerates disguised as innovation platforms, issuing debt at record-tight spreads (74bps on average) as if cash flow risk were a rounding error. Yet many are funding multi-decade energy and infrastructure projects with three-year paper. In trader terms, the street’s running an unhedged carry trade on the future of AI.

AI is a power story before it’s a productivity story. Each new data center requires gigawatts of electricity and millions of gallons of cooling water. Tech/AI analysts peg the capital need at half a trillion dollars a year — equivalent to rebuilding the global energy grid every decade. The irony is breathtaking: the technology meant to optimize the world’s resources is consuming them at a pace that could destabilize credit markets.

And like all leverage-driven manias, the narrative hides the fragility beneath. If even one leg of this scaffolding falters — say, a breakthrough in quantum computing that renders current GPUs obsolete, or a China-built LLM that undercuts Western margins — the equity bubble will deflate. But the real damage will come from the credit side: the loans, securitizations, and private notes written on top of revenue projections that never arrive.

AI’s great paradox is that it has become the world’s most capital-intensive idea. We’ve turned “intelligence” into an industrial commodity, financed like shale or steel. The market is now pricing intelligence as if it were oil — barrel by barrel, watt by watt — ignoring that intelligence, unlike oil, doesn’t always pay for its own extraction. And that’s where the risk lies. The credit market has effectively securitized hope. Insurance funds and sovereigns are holding paper backed by projected AI revenues — not realized profits — in the same way subprime lenders once packaged NINJA loans as AAA tranches. Only this time, the collateral isn’t a house; it’s a GPU cluster in Nevada.

We’ve entered the AI-as-debt era — where intelligence itself is being leveraged, packaged, and sold as a financial product. This isn’t to say the AI dream will implode. It may evolve. Technological breakthroughs could save it — more efficient ASIC chips, new cooling systems, or genuine quantum computing leaps that cut costs. But the math is stubborn: unless productivity growth catches up, the sector is borrowing its way into an illusion of progress.

When future cash is discounted back to the present at too low a rate, you don’t get innovation — you get asset inflation. Every generation’s “new economy” starts with promises and ends with leverage. The question is no longer whether AI will change the world — it’s whether the world can afford the debt required to find out.