AI Data Centre Debt Splits as Pimco Warns Investors

The AI boom now has a debt problem hiding behind the glamour.

Data centres need land, power, chips, cooling, and huge loans. That money is rushing in fast. A year ago, high-yield debt tied to data centres was barely a market. Now it forms about 4 percent of the high-yield market, and could reach 10 percent in two years.

For Indian investors, this is not some faraway Wall Street curiosity. If your mutual fund owns global debt, tech stocks, or AI-linked themes, this plumbing matters.

Data centre debt is splitting

Pimco executive David Forgash has warned investors to tread carefully in data centre debt. He said winners and losers have already started to appear.

High-yield debt means borrowing by companies that pay higher interest because lenders see more risk. Traders often call it junk debt, though not every borrower is weak.

The key number here is the spread. A spread is the extra interest a borrower pays over safer government bonds.

Forgash said about 75 percent of this data centre debt trades at spreads near 6 percent. The remaining 25 percent trades at 10 percent or more.

That gap tells you something simple. The market trusts some projects. It worries deeply about others.

AI demand is driving borrowing

The rush comes from artificial intelligence. AI needs vast computing power, and that power sits inside data centres.

Large technology companies, often called hyperscalers, have borrowed more than $250 billion globally for this infrastructure. In the US high-yield market alone, data centre borrowing now sits around $40 billion in face value.

That number has grown mostly within the past year. Speed is the risk here.

Markets love a new theme when money is easy. Then they discover boring questions. Who has signed long contracts? Who pays the electricity bill? Who gets stuck if AI demand slows?

Forgash said Pimco prefers deals backed by stronger companies. It also wants contracts that customers cannot easily cancel.

That sounds dull, but it matters. If a tenant can walk away, the lender may be left with a costly building and no steady cash flow.

Why Indian investors should care

Most Indian households will not directly buy US data centre bonds. But risk travels through funds, stocks, and sentiment.

A retail investor with a ₹5 lakh portfolio may own international funds, tech funds, or debt schemes with global exposure. If investors suddenly mark down risky AI-linked debt, those losses can ripple into fund values.

The connection may not appear obvious on day one. But market stress rarely stays in one neat box.

We saw this with real estate credit, telecom debt, and shadow lenders in different markets. The first warning usually looks technical. Then investors realise it affects real money.

For Indian tech workers, the other link is jobs. Forgash said AI disruption is hitting software and services. That matters because India has built a large export engine around software work.

He added that software forms about 15 percent of leveraged loans, but only 3 percent of high-yield bonds. In plain English, risky loans have more software exposure than risky bonds.

That limits damage in one corner of the bond market. It does not remove pressure on the software business itself.

Contracts matter more than hype

The cleanest data centre story has three parts.

First, a strong customer signs a long contract. Second, the project earns enough cash before the contract ends. Third, the building still has value after that.

Pimco is looking for that structure. It wants repayment before contracts expire, and some leftover value after the main deal ends.

That is the difference between infrastructure and speculation. Infrastructure throws off steady cash. Speculation depends on the next buyer paying more.

Investors often forget this during a boom. They see AI demand and assume every data centre will fill up.

But data centres are not all equal. Power access matters. Location matters. Cooling costs matter. The quality of the tenant matters most.

The market is calm, but selective

Forgash’s warning is not that the entire data centre debt market is broken. His point is sharper.

The calm headline numbers hide stress inside the market. Three-fourths of the debt looks manageable. One-fourth already demands much higher interest.

That is why the split matters. A 6 percent spread says investors see risk but can live with it. A 10 percent spread says they want serious compensation.

For borrowers, that higher cost can change the whole project. More interest means less room for delays, power cost shocks, or weaker demand.

For lenders, it means the market has started asking harder questions. That is usually healthy, but it can hurt late entrants.

Goldman Sachs Group hosted the credit conference where Forgash discussed these risks. The setting itself says plenty. Data centre debt has moved from niche financing to a serious credit market debate.

Pimco also sees opportunities outside this crowded AI trade. Forgash pointed to homebuilders and building suppliers, which have sold off even though future rate cuts could help them.

He also said the firm remains overweight on energy through year-end. That is another reminder that not every market story begins and ends with AI.

For ordinary investors, the lesson is simple. AI may transform business, but borrowed money still follows old rules. Cash flows matter. Contracts matter. Price matters. When a market grows from almost nothing to billions in one year, the question is not whether the story is exciting. The question is who gets paid back when excitement meets interest costs.