You spent the last two years deciding whether to buy the AI trade. While you deliberated, the bond fund inside your retirement account bought it for you. Not a little of it. Mechanically, automatically, and in size — with no prospectus disclosure that said “AI infrastructure lender.”
This is the part of the AI boom nobody put on a chart. The equity story is exhausting and well covered: capex is enormous, valuations are stretched, Nvidia is 13% off its high on circular-financing worries. Fine. But the AI buildout has quietly changed how it gets funded, and that change moved the risk from the side of your portfolio you watch to the side you don’t.
AI debt stopped being a rounding error
Through 2024, hyperscalers paid for AI with operating cash flow. They had staggering amounts of it, and the story was simple: Microsoft, Alphabet, Amazon and Meta were the most self-funding businesses in corporate history.
That ended. The five largest operators have guided to roughly $775–800 billion in 2026 capex, about three times the ~$238 billion deployed in 2024. Cash flow, enormous as it is, does not stretch that far. So the bond market got the call.
- Five major hyperscalers issued about $121 billion of US corporate bonds across all of 2025.
- Six of the biggest tech names issued roughly $244 billion globally through mid-July 2026 alone.
- Morgan Stanley forecasts AI-linked debt issuance tops $570 billion in 2026 — close to a doubling.
- Goldman Sachs expects Big Tech to fund more than a third of its AI investment with debt by 2027.
AI-related investment already accounted for roughly 30% of total net issuance in the US dollar investment-grade market in 2025. Total IG supply is projected at $2.46 trillion in 2026, up 11.8% — with AI capex named as the primary driver of the increase.
Read that sequence again. In under two years, the AI buildout went from being financed by the most profitable companies on earth to being financed by whoever buys investment-grade bonds. That is a different risk with the same ticker symbols on top of it.
The mechanical part: index funds have no opinion
Here is the piece that should make you sit up, because it requires no forecast and no view on AI at all. It is arithmetic.
Major corporate bond indexes are weighted by the amount of debt outstanding. Every index-eligible bond a company issues increases that company’s weight in the index. Every passive fund benchmarked to that index must then hold proportionally more of it. There is no committee, no conviction, no “do we like the AI capex-to-revenue gap here.” The bond exists, so the fund buys it.
The tech sector is now about 10% of the Bloomberg Corporate Bond Index, up from 9% in 2024. That sounds trivial. It is not the number that matters. Vontobel estimates roughly $300 billion of AI and data-center-related issuance over the next year and about $1.5 trillion over five years, which would take the AI-related segment to 15–20% of most corporate bond indexes.
Now connect it to where ordinary money actually sits. Target-date funds held about $4.8 trillion at the end of 2025. Those funds hold bond index funds. Which means tens of millions of people whose entire financial plan is “contribute to the 401(k) and don’t touch it” are becoming AI infrastructure creditors on a glide path — increasingly so as they age into a higher bond allocation.
The cruel symmetry: the closer you are to retirement, the more of your portfolio sits in the bond sleeve, and the more of that sleeve is becoming AI debt. The de-risking mechanism is quietly re-risking.
The contrarian part: watch spreads, not earnings
Every retail investor watches the same two things — Nvidia’s earnings and hyperscaler capex guidance. Both are lagging indicators of a story that has already moved to a different market.
When a buildout is funded from cash flow, the constraint is willingness: a CFO decides how much to spend. When it is funded from the bond market, the constraint is access: someone else decides whether to lend, and at what price. That is a much harder constraint, and it tightens faster.
It is already tightening. Goldman’s AI bond basket has seen spreads widen sharply, and the market’s absorption capacity has visibly shrunk — where roughly $75 billion of supply once stressed the market, about $25 billion now puts it on the defensive. Credit default swap spreads on AI-linked names widened well beyond the broader market. Meta is reportedly paying more for its latest data center financing than it did on its record 2025 deal, and is arranging roughly $12 billion through a BlackRock-backed special-purpose vehicle for a near-gigawatt site in El Paso. Oracle has raised about $25 billion. Spreads on new Amazon and SpaceX paper widened in secondary trading.
And this is happening into a long end that is already expensive — as we wrote when the 30-year hit 5.21% while inflation was falling, the term premium repriced for reasons that had nothing to do with CPI. Add $570 billion of AI supply to that and you are not asking the bond market a small question.
So the thing that ends this cycle probably isn’t a bad quarter at Nvidia. It’s a failed or badly-priced deal — a hyperscaler that has to pay 40 basis points more than it planned, or pulls a tranche. Equity investors will find out about it second.
The part that isn’t on any balance sheet
There is more leverage here than the bond totals show. Moody’s flagged roughly $662 billion of data center lease commitments signed but not yet commenced, which sit off balance sheet under GAAP’s lease-commencement rules. That figure is larger than the combined on-balance-sheet debt of the same companies.
Pair that with the accounting timing problem we covered in the $200 billion line nobody reads — capex recognized as expense over five and six year schedules, so today’s income statement carries a fraction of today’s spend — and the picture is consistent. Obligations are being taken on faster than they are being reported, and the reporting lag flatters exactly the metrics people use to justify the next round.
None of this means sell
The buildout is real and the demand behind it is real. Data centers get built, models get better, and a large amount of genuine value gets created — this is not a pets.com situation where the underlying thing doesn’t work. Debt-funding a productive asset is what companies are supposed to do.
The point is narrower and more useful: you no longer get to opt out by asset class. The old mental model — equities are the risky sleeve, bonds are the safe sleeve — assumed the two sleeves had different exposures. In an AI buildout funded by investment-grade issuance, the same handful of balance sheets sit on both sides of your portfolio. Diversification that runs across asset classes but concentrates in the same five borrowers is not diversification. It’s the same bet, priced twice.
What actually to do about it
- Open your bond fund’s holdings. Not the equity fund — the bond fund. Look at the top 10 issuers. If four of them are hyperscalers, you now know something you didn’t know this morning.
- Count exposure by borrower, not by asset class. Add your equity weight in the megacaps to your credit exposure to the same names. That total is your real concentration — the same lesson as the Nasdaq’s 10% fall against the S&P’s 4%.
- Treat IG credit spreads as an AI indicator. They are free to watch and they lead the equity narrative.
- If you want bond exposure that isn’t an AI bet, Treasuries and short-duration government paper carry none of this issuer concentration. That is a genuine choice you can make deliberately instead of by default.
The wealthy have always understood something the saving class hasn’t: risk is not organized by product category, it’s organized by counterparty. You can own six funds and still be exposed to five companies. The AI buildout just made that true for people who never chose it.
Want to track your own path to financial independence? The Wealtharian Wealth Tracker lets you monitor your net worth, FU money progress, and investment milestones in one place. Try it free →