The Bet: AI Gets Its Own Credit Bubble

My call: the world will be sitting on more than $2 trillion of AI‑linked debt by the end of 2027. Not notional marketing fluff, but outstanding bonds, loans, and structured deals that at least two serious data shops agree to count as “AI credit.”

The consensus comfort object says AI is an equity story and Big Tech balance sheets are pillows of cash. The signal says something else. Hyperscalers are quietly turning AI into a credit supercycle, and the bond market is too flattered to complain.

If this is right, we are not just funding a productivity boom. We are building the scaffolding for the first trillion‑dollar theme bubble that will never dare call itself one.

The Drivers: GPUs On Layaway

Start with the obvious: data centers, GPUs, and power are now public utilities with better branding. Somebody has to pay the trillions of dollars of upfront capex. Shareholders like the AI story, they like it less when you issue stock. So CFOs do what CFOs do. They call the debt desk.

Big Tech is already more than 8 percent of the US corporate bond market, a record high. Amazon is lobbing in surprise bond deals in the mid‑tens of billions, specifically to feed its AI infrastructure habit. Microsoft, Alphabet, Meta and friends are on similar trajectories. This is the spine of the coming stack: investment‑grade credit underwriting the GPU grid.

Banks see it. Bank of America says it has helped raise nearly $500 billion for AI‑related companies since 2025 across investment‑grade, leveraged finance, and equity. That number is not a think‑piece projection. That is fee revenue.

On the allocator side, Singapore’s Temasek has already put AI in the org chart. It plans to lift AI‑related investments from 6 percent of its SGD 518 billion portfolio to 15 percent by 2031 and more than double private credit from 2 to 5 percent. You do not write that down in a policy deck unless you intend to hand managers multi‑year tickets labeled “AI” and “private credit” and tell them to stop underperforming the MSCI World.

That is the core loop. Big Tech wants AI capex without equity dilution, banks want fees, allocators want “equity‑like returns with lower risk,” and private credit funds are happy to warehouse the stuff banks do not want to hold. Call it the GPU carry trade.

From Equity Story To Credit Supercycle

The line from “cool demo” to “ABS tranche” is shorter than it looks.

AI platforms are printing revenue slides that would have made late‑90s dot‑coms blush. Anthropic, Sierra, Glean, AI‑integrated incumbents like Gusto: all telling the market that annual recurring revenue ramps are vertical and churn is theoretical. Whether you believe those curves is almost secondary. In modern markets, believable PowerPoint is collateral.

JPMorgan is already talking about AI‑linked assets as a potential number‑two asset‑backed securities market after mortgages. SemiAnalysis has a publicly circulated scenario for $7 trillion of AI debt by 2029. These are not wild Reddit threads, they are the slideware that product committees use to justify building AI‑labeled securitizations.

Expect the pattern we have seen before: first corporate bonds and term loans tied to named AI spend, then project finance for data centers and compute clusters, then increasingly baroque structures backed by AI usage fees, enterprise subscriptions, and GPU leasing cash flows. Once the legal templates exist, selling a new theme is mostly a matter of changing the cover and booking the roadshow.

By 2027, a $2 trillion outstanding stack is not heroic. It is the midpoint between a known starting point and the sell‑side’s own bullish slide packs.

The Pushback: What If The Future Underperforms?

Not everyone is drunk on the Kool‑GPU. Vishal Khanduja at Morgan Stanley is already on TV saying the quiet part: Big Tech credit risk looks too cheap, and the AI debt binge is flashing early warning lights. He called Amazon’s surprise $25 billion bond sale just that, a surprise, in a market that thought the tab was already tall enough.

There are real failure modes here.

• Rates stay higher for longer, turning long‑duration AI capex into a negative NPV hobby.
• AI revenues disappoint once enterprise pilots run into messy reality and IT budgets, not slideware, set the pace.
• Regulators decide that infinite data centers are bad for grids, privacy, or labor politics and put a ceiling on the rollout.

Any one of those could slow issuance or reprice spreads. Enough of them together could freeze the market before it crosses the $2 trillion line.

There is also the classification mess. “AI‑linked” is conveniently vague. Is Nvidia’s entire bond stack AI credit? What about a cloud provider that spends half its capex on AI and half on everything else? If data vendors and banks punt on clean tagging, we might have a multi‑trillion AI credit stack that officially does not exist.

My forecast builds in those risks and still comes out bullish. Not because the assets are obviously safe, but because the incentives to ignore that question are stronger than the incentives to slow down.

The Stakes: Productivity Or Prettied‑Up Leverage

If the call is right, by late 2027 AI will not just be a line item in tech portfolios. It will be a defined credit vertical, tracked by Bloomberg fields and ETF tickers, with banks marketing AI ABS to anyone who missed the equity rally.

The winners in that world are obvious: hyperscalers that funded the build‑out at yesterday’s tight spreads, banks that got years of fees, and allocators who rode the wave before spreads normalized. The losers are still abstract. They are whoever is left holding long‑duration AI paper if revenue curves flatten and regulators rediscover their spine.

This desk will score the call cleanly. If we do not see at least two independent estimates of AI‑linked outstanding debt converging around $2 trillion by December 31, 2027, chalk it up as wrong and send the hate mail. If we do, remember how many people will insist they always knew this was a sober, prudent, productivity‑financing exercise.

After all, nothing says responsible innovation like turning the end of history into a buy‑now‑pay‑later plan.