About $300 billion of debt tied to AI projects has flooded credit markets and investors are starting to get picky. Big tech borrowers have leaned on bonds and loans to fund data centres and AI hardware, with offers from Meta, Alphabet, Oracle and others drawing large but cooler demand than before. Bankers are adding incentives and lenders are asking for stronger protections as supply keeps rising and deal terms evolve. The shift means credit markets are now sorting which AI bets belong on balance sheets and which should stay with equity.

Credit desks fast-tracked deals last year and into 2026. Lenders answered with hundreds of billions in bond and loan purchases. The wave covered investment-grade bonds, leveraged loans and riskier project financings. Now buyers want clearer protection for the money they tie up.

Big names led the push. Oracle tapped the bond market for about $18 billion in September. Alphabet sold roughly $20 billion of debt, including a rare 100-year sterling bond. Meta ran a jumbo bond sale that drew a peak order book near $96 billion for a deal expected to raise as much as $25 billion. Those moves helped create the roughly $300 billion tally that market participants now point to.

Why tech is borrowing

Hyperscalers have pushed capex higher to build out AI compute. UBS analysts estimate hyperscaler capital spending could top $770 billion in 2026, about 23% above previous expectations. UBS credit strategists said that rise implies an extra $40 billion to $50 billion of borrowing, and could lift public market debt issuance into the $230 billion to $240 billion range this year.

The math is simple. AI hardware, data centres and networking cost a lot. Companies want to keep growing capacity before revenue from AI fully materialises. So they're using debt markets to bridge the gap between investment and future sales. That choice has changed how investors view these tech balance sheets.

Signs of investor fatigue

Deals still clear. But bankers are doing more to sell them. That includes higher fees and extra concessions to investors.

Underwriters are offering better economics to lure buyers who now have plenty of options.

Demand for some recent sales has been lower than earlier deals. Meta's recent bond drew less demand than a sale in October, when orders reached about $125 billion for a $30 billion deal. Market participants say the difference shows appetite is cooling even as headline interest remains.

"these companies are selling a lot of debt and they're going to have to pay up to borrow," said Robert Tipp, head of global bonds at PGIM Fixed Income. "The market, after a spectacular narrowing in corporate spreads to historical tights, is seeing a wall of worry piled up before it."

Mastheads

Deal terms are changing

Investors are asking for more protections. Borrowers are agreeing to amortisation, so some principal gets repaid before final maturity. That reduces long-term exposure for lenders. Other protections are appearing too.

In riskier corners, issuers are getting credit backstops from hyperscalers. Those covenants promise that a data-centre lease will be paid even if a tenant defaults. Some deals now include cost ceilings or clauses that trim owners' exposure to runaway construction bills.

"We're seeing what different investors value when it comes to these financings and how they're evaluating risk and return," said John Servidea, global co-head of investment-grade debt capital markets at JPMorgan Chase & Co. "We're seeing really good demand for these deals but as supply increases, we expect deal terms and structures to continue to evolve."

What lenders are asking

Portfolio managers are running through risk checklists. They want to know whether projects face execution problems. They ask about supply chain and construction delays. The group check tenant quality and contract length for data-centre deals.

"Do you have execution issues? Supply chain or construction delays? Is it a lower-quality tenant?" said David Kinsley, senior portfolio manager at Impax Asset Management. He said investors are getting better at spotting project-level risk and are turning away from deals that fail those checks.

That scrutiny is most intense where leverage is higher. The riskier parts of the market don't have long-standing conventions for how protective clauses translate into yields. Money managers told bankers they're still figuring out fair compensation for specific protections.

Borrowers get cheaper capital when demand is strong. That helped hyperscalers until now. But issuing more paper pushes supply into markets that are already digesting big government deficits and corporate debt. Asset managers warn that rising corporate borrowing adds to supply pressure.

Investors who bought early-stage data-centre or AI-related paper stand to gain if those projects start generating revenue as hoped. Bond buyers who enter later face different choices. They can demand higher yields, tighter covenants, or both. Or they can wait.

Al Cattermole, fixed income portfolio manager at Mirabaud Asset Management, said the shift is changing how the market treats giant tech names. "For years, we've been told this AI spend would be funded by generated cash flow, that it's equity risk, it's speculative, and not to worry about it from a credit point of view," he said. "There now seems to be a change in the unspoken contract that while we would continue to lend to these businesses, really AI capex was still going to be equity or cash funded."

Markets are balancing two facts. One, companies need capital to scale AI. Two, lenders want protection for that capital. The contest shows up in prices and docs. Spreads tightened markedly as demand surged. Now they're under pressure to widen or for deals to offer more protective covenants.

Bankers say demand remains solid. But they're also realistic that more supply will force changes. "We're seeing really good demand for these deals but as supply increases, we expect deal terms and structures to continue to evolve," John Servidea said.

Related Articles

Meta's recent bond attracted a peak order book of about $96 billion for a sale expected to raise up to $25 billion.

This article was created with AI assistance.