AI data center debt has moved from a financing detail to a credit-market warning sign, and nearly half of surveyed fund managers now see hyperscaler capex as the likeliest source of a systemic credit event.
The private placement market used to be dull in the best possible way. Railroads, canals, mines, utilities, and investment-grade industrials sold debt directly to institutions that wanted long assets and few surprises. Now it is financing AI data centers. That is new.
Bloomberg reported this month that private bond issuance reached roughly $81 billion through May 2026, the strongest start in data going back to 2016, citing Private Placement Monitor. The same Bloomberg account said AI infrastructure is helping drive the surge, as technology borrowers look for capital and life insurers look for long-dated bonds to match annuity liabilities. You should pay attention to that pairing. It tells you the AI buildout is no longer only a stock-market story. It has become an insurance balance-sheet story too.
IREN Ltd. is a clean example of how fast the market has changed. The company, previously known for bitcoin mining and now building AI cloud infrastructure, announced on June 1 that it had closed a $3.65 billion investment-grade GPU financing facility for its Microsoft contract. The package included a $2.10 billion U.S. private placement at a fixed rate equal to SOFR plus 2.13%, alongside a $1.55 billion delayed-draw term loan. IREN said the deal was secured against GPUs and contracted cash flows, and that Goldman Sachs and J.P. Morgan arranged it with participation from insurers, asset managers, and other institutions.
That is not a small correction.
The earlier version of this story described IREN as selling roughly $2.1 billion of bonds in May. The verified record is more specific: Bloomberg reported in May that Nvidia agreed to invest up to $2.1 billion through purchase rights tied to a commercial agreement, while IREN separately closed the $3.65 billion GPU financing in June. Those are different transactions. Mixing them together makes the story look tidier than the credit market really is.
The debt boom is real
The bigger numbers are just as hard to ignore. Reuters reported that Morgan Stanley expects global AI-related debt issuance to reach nearly $570 billion in 2026, with about $236 billion raised by May 31. That was roughly four times the level from the same period a year earlier. Reuters also said Morgan Stanley expects hyperscaler capex to exceed $1 trillion in 2027, with Alphabet, Amazon, Microsoft, and Meta expected to spend about $700 billion this year.
Here's the thing: those companies aren't weak borrowers. Amazon, Microsoft, Alphabet, and Meta still have real earnings power, large cash piles, and public markets that will lend to them. The stress is more awkward than that. A growing share of AI infrastructure financing is being pushed into project-style structures, private placements, asset-backed deals, and credit funds where the repayment story depends on power, construction, tenant commitments, GPU values, and timing. You don't need a collapse in AI demand for that to bite. A delay can be enough.
Forbes recently noted that bond demand has softened as AI-related supply has grown, with investors paying closer attention to coverage ratios, spreads, and whether deals are backed by a corporate borrower or by a narrower pool of data center assets. That distinction matters. A direct claim on Amazon is not the same thing as a claim on one leased building or a stack of GPUs that may depreciate faster than the debt matures.
PIMCO has been blunt about the wider credit backdrop. In its 2026 secular outlook, Richard Clarida, Andrew Balls, and Daniel Ivascyn wrote that the "default cycle is reasserting itself," and said the firm expects significantly higher losses in lower-quality credit such as leveraged and private direct lending. The firm also warned that financial engineering is accelerating in private credit, insurance balance sheets and other specialised structures, while highly engineered financings tied to AI deserve careful analysis.
That sounds cautious. It should.
Investors are nervous and still buying
Bank of America's July Global Fund Manager Survey shows the contradiction clearly. Benzinga reported that 48% of investors named AI hyperscaler capex as the most likely source of a systemic credit event, ahead of private credit and consumer credit. The same survey found that 45% called the AI bubble the biggest tail risk, up from 28% in June. Long global semiconductors was still named the world's most crowded trade by 82% of respondents.
They are trimming, not leaving.
Only 28% of surveyed managers expected a hyperscaler to announce a capex cut this year, according to Benzinga's readout of the BofA survey. Newsquawk's summary of the same July survey said 61% did not expect a hyperscaler capex cut. However you slice the wording, the message is plain enough: investors fear the borrowing boom, but most still expect the spending machine to keep running.
That is the real risk for startups and infrastructure operators trying to fund the next wave of AI capacity. The cheap debt window is narrowing. Life insurers and private credit funds are still there, but they are starting to demand more yield, stronger collateral, and documentation that does more than wave at a hyperscaler logo - cleaner tenant commitments included. If you're raising non-dilutive capital now, the buyer across the table is no longer paying for the mood of 2024.
Fresh news from IREN shows why this story is still moving. MarketWatch reported this week that IREN shares jumped after the company announced $2.8 billion of new multiyear AI cloud-service contracts and raised its 2026 annualized AI cloud revenue target to $4 billion. That helps the equity story. It also proves the credit story is alive, because every new contract can support more financing, more equipment to buy, and a bigger leverage stack on top.
The private placement market can absorb a lot. It has done that for more than a century. But AI data centers bring a different set of risks from railroads and utilities: shorter technology cycles, power bottlenecks, concentrated tenants, and hardware that can lose value quickly if the next chip generation arrives faster than expected. A generational infrastructure buildout may still be exactly that. It can also produce a painful credit cycle before the winners are clear.
Also read: Tencent stock falls 7% as gaming fears and an AI rotation hammer China tech • Big Tech is spending $725 billion on AI in 2026 and its free cash flow is nearly gone • BlackRock's IBIT posts longest Bitcoin ETF inflow streak since April as CLARITY Act optimism returns
