JPMorgan sounds the alarm on the runaway cost of AI | Official Website of Louis Velazquez, entrepreneur, finance guy and tech innovator

Every era has a number that stops making sense, a figure so large the brain just files it next to “infinity” and moves on. For most of my career, that kind of number lived in government budgets, the sort of sum you nod at without really feeling.

Lately it has migrated somewhere stranger. It now sits in a single stock ticker, in the valuation of a company that makes chips.

Nvidia (NVDA) is worth about $5.5 trillion. To put that where you can feel it, the chipmaker is now larger than the annual economic output of every nation on Earth except the United States and China, according to 24/7 Wall St. It overtook Germany’s economy in mid-May 2026, according to Euronews.

We have watched that number climb, and the worry has pointed at one place, the stock. Is it a bubble? Will the AI trade unwind and drag our retirement accounts down with it?

Fair questions. They are also, by my read, aimed slightly off target. Because the warning that grabbed me this spring did not come from a short seller. It came from a bank, and it was not about the stock. It was about the debt.

JPMorgan flags a concentration risk in the corporate bond market as AI borrowing climbs.

fatido / Getty Images

Why the AI boom got too big to ignore

To understand why a bank is nervous, start with how much money the AI build-out actually requires. The five companies doing most of the spending, Alphabet (GOOGL), Amazon (AMZN), Meta (META), Microsoft (MSFT), and Oracle (ORCL), have committed roughly $969 billion combined to data centers and the chips inside them, according to Fortune.

More AI:

For years, that spending came straight out of operating cash flow. That is changing. The build-out now costs more than even these giants generate, so they have started borrowing the difference.

In 2025, those five firms issued about $121 billion in U.S. corporate bonds, up from an average of $28 billion a year between 2020 and 2024, according to Reuters. That is not a rounding error. It is a structural shift in who feeds the bond market.

What JPMorgan sees building inside the bond market

Here is the part that should make you sit up. As of late May 2026, debt tied to AI made up roughly 15% of the entire corporate bond universe, a share one Morningstar host called “very high by historical standards,” according to 24/7 Wall St.

The bigger worry is not that these companies are over-borrowed. On most balance-sheet measures, they remain very strong, according to J.P. Morgan Asset Management. The worry is concentration.

Technology has grown from less than 2% of the investment-grade credit market in 2005 to about 10% today, the same firm found, and credit spreads in the sector have widened as investors digest the flood of new paper. The risk has quietly moved from any single company to the shape of the market itself.

The scale is hard to hold in your head, so here are the numbers that matter:

The five hyperscalers issued about $121 billion in bonds in 2025, versus a $28 billion annual average from 2020 to 2024, according to Reuters.
AI-related debt now accounts for roughly 15% of the entire corporate bond universe, according to 24/7 Wall St.
Technology has climbed from under 2% of the investment-grade credit market in 2005 to about 10% now, according to J.P. Morgan Asset Management.
Wall Street expects another $100 billion to $300 billion of AI-related bond supply in 2026, according to Fortune.

The trajectory is what alarms JPMorgan. The bank projects that by 2027, spending on AI will outrun the entire world’s military budget, which reached $2.718 trillion in 2024, according to the Stockholm International Peace Research Institute. Sit with that. We may be one year away from the world borrowing more to build chatbots and data centers than it spends to arm itself.

How a bond market bet lands in your portfolio

Here is why this matters even if you never buy a single corporate bond. If you hold a total-bond-market index fund, and millions of retirement savers do, you already own this AI debt. Index funds have to mirror the market, so as hyperscalers flood it with bonds, your fund quietly fills with their paper whether you wanted that bet or not.

Professional buyers feel the squeeze, too. Kevin SigRist, who runs North Carolina’s $143 billion pension fund, said the extra yield for holding these long bonds is thin, with spreads “very, very tight,” according to Fortune. His fund stays underweight the sector on purpose.

When I ran the comparison myself, the thing that stuck was not the size of any one company. It was the wiring. A single theme, AI, now sits inside your stock funds through the Magnificent Seven and inside your bond funds through hyperscaler debt. It is the same bet, made twice, in the two halves of a portfolio that are supposed to offset each other.

That is the cost JPMorgan is really flagging. Not a weak balance sheet at any one firm, but a market that has quietly pointed most of its money at one story.

The fix is not to panic-sell or to swear off AI. It is to know what you actually own. Pull up your largest holding, in your 401(k) or your brokerage, and check how much of it rides on five companies. If the AI story keeps compounding, you will be glad you looked. If it stumbles, you will be very glad you looked. Either way, the number that stopped making sense is now sitting in your account. The least you can do is learn its name.