Morgan Stanley Leads the AI Debt Boom as Ratepayers Absorb the Cost

Morgan Stanley expects companies building AI infrastructure to issue $570 billion in debt this year, and the bank wants to arrange more of it than any rival on Wall Street. By May 31, borrowers had already sold $236 billion of that debt, roughly four times the pace of the same stretch in 2025.

Buried under that headline number is a quieter one. Electricity ratepayers, state budgets and Wall Street’s own risk surveys are already absorbing pieces of this borrowing spree that rarely make it into the forecast.

Morgan Stanley’s Ledger Fills Up

The bank is not only forecasting the boom. It is arranging it.

In September 2025, Morgan Stanley put together a $3 billion debt facility for TeraWulf, a Bitcoin mining company that has pivoted into AI and data center infrastructure, with backing from Google. The deal is one of the clearest examples yet of a crypto-era balance sheet being rebuilt to carry AI-scale debt.

A Bond Market Bigger Than the Banks

Hyperscaler investment-grade bond issuance, meaning debt sold by the massive cloud and AI platform operators, surpassed $100 billion in 2025. Morgan Stanley expects net supply to climb another 30 to 50 percent in 2026, landing somewhere between $130 billion and $150 billion.

The pool is already large enough to matter to anyone holding a bond fund. By October 2025, AI-linked debt had reached $1.2 trillion outstanding, the largest single segment of the investment-grade market and bigger than the U.S. banking sector’s share of the same benchmark, according to data from M&G Investments.

Morgan Stanley and TeraWulf are not the only names filling that pipeline.

• Hut 8 closed a $4.25 billion bond on June 10, 2026 to fund its Beacon Point AI data center in Texas.
• Keel Infrastructure closed a $458 million convertible notes deal a day earlier for its own data center buildout.
• Core Scientific, another former crypto miner turned data center operator, has expanded its existing credit facilities to keep pace with the same demand.

Each of these borrowers is smaller than the hyperscalers driving the headline numbers, but together they show how far down the AI supply chain the borrowing has already spread.

Who Actually Pays for the Power

Data centers do not run on debt alone. They run on electricity, and someone has to build the plants and lines to deliver it. U.S. investment-grade utility issuance hit approximately $135 billion in 2025, and Morgan Stanley expects that to climb to around $145 billion in 2026, driven largely by the need to power new data center capacity.

Data centers have already hiked electricity prices on the public by $23 billion, Fortune reported, and how that cost will ultimately be split between tech firms and everyone else on the grid is still unresolved.

It’s just so far beyond any bill that I’ve ever had.

John Steinbach, a longtime homeowner in Manassas, Virginia, said that about the $281 electricity bill he received in January 2026, more than double the roughly $100 he paid the month before. His experience is not isolated. State lawmakers introduced more than 300 bills on data center costs during 2026 alone, according to Consumer Reports.

Why Are States Pumping the Brakes on Data Centers?

At least 14 states are weighing moratoriums on new data center projects in 2026, Moody’s Ratings found, worried that the fiscal and grid strain of hosting hyperscale AI facilities is outpacing the tax revenue and jobs they bring. Michigan and West Virginia are moving the opposite way, using incentives to pull the same investment toward their regions.

Moody’s drew a direct line in June between the data center boom and elevated fiscal pressure on state and local governments, warning that the public power and water systems those projects lean on were not built with hyperscale AI workloads in mind.

Maryland Democratic state Del. Lorig Charkoudian said the mismatch between projected and actual data center demand is not abstract. “It actually has a very concrete financial impact on ratepayers,” she said, describing a wave of new state data center rate laws now moving through statehouses. Georgia’s Public Service Commission already requires data centers to fund the upstream generation, transmission and distribution costs their load creates, and Oregon passed a similar law carving out a separate rate class for large, long-term data center contracts.

The Debt Wall Street Keeps Off the Books

Not all of this borrowing shows up where investors can see it. Moody’s Ratings calculates that Amazon, Meta, Alphabet, Microsoft and Oracle had amassed $969 billion in future data center lease commitments by the end of 2025. Of that, $662 billion covers leases that have not yet begun, meaning the obligation sits entirely off the five companies’ balance sheets for now.

Moody’s analyst David Gonzales described that pile as “yet to be on the balance sheet” rather than a hidden liability, in an accounting analysis of hyperscaler lease obligations, though the total equals 113 percent of the five companies’ most recent adjusted debt.

The disagreement over how to treat that debt has split the two biggest rating agencies. Last October, S&P let Meta keep the cost of its Hyperion data center in Louisiana off its balance sheet entirely, allowing the company to raise $27.3 billion through Beignet Investor, a special purpose vehicle it owns jointly with the private credit firm Blue Owl Capital. Moody’s pushed back, warning that “we foresee a material increase in adjusted debt and lease-related cash outflows for these companies,” analysts Gonzales and Alastair Drake wrote in an opinion challenging the accounting treatment banks are using for these deals.

• Moody’s Ratings says off-balance-sheet lease deals understate real leverage and plans to apply its own debt adjustments when rating hyperscalers that use them.
• S&P Global Ratings allowed Meta to keep $27.3 billion of Hyperion financing off its balance sheet through the Beignet Investor structure.

Fund Managers’ Fear Is Outrunning the Debt Itself

Wall Street’s own investors are getting nervous faster than the debt is piling up. A Bank of America survey from May 2026 found that 34 percent of global fund managers identified AI hyperscaler capital spending as the most likely source of a future systemic credit event, double the share who said so in April, Bloomberg reported.

By July, that number had jumped again. In Bank of America’s July 2 to 9 survey of 210 fund managers overseeing $555 billion, 48 percent picked AI hyperscaler capital spending as the likeliest trigger for a systemic credit event, ahead of private credit’s 34 percent for the first time. The “AI bubble” itself became the single biggest tail risk in the same survey, cited by 45 percent of respondents, up from 28 percent in June.

Private credit topped that worry list two months earlier. AI infrastructure spending has since taken over.

TeraWulf’s Pivot Shows How the Risk Stacks

The company at the center of Morgan Stanley’s marquee AI deal was not built for this business. TeraWulf started as a Bitcoin mining operation, competing on cheap power and hash rate, before repositioning its data centers to host AI workloads.

TeraWulf’s data centers were mining Bitcoin as recently as a few years ago. Today they carry a Google-backed, Morgan Stanley-arranged loan built for an entirely different industry, one more sign that this year’s $570 billion is financing infrastructure that has already changed hands once before.

Frequently Asked Questions

What counts as AI-related debt in Morgan Stanley’s forecast?

Morgan Stanley’s $570 billion figure covers debt raised globally by hyperscalers such as Amazon, Alphabet, Meta, Microsoft and Oracle, plus semiconductor makers and data center developers, sold in both dollar and non-dollar bond markets.

Why are hyperscalers raising equity as well as debt?

Bond markets alone have not closed the gap between AI capital spending and operating cash flow. Google completed an $84.75 billion equity raise in 2026 that included mandatory convertible preferred shares, and analysts at Barclays expect Meta, Microsoft and Amazon to consider similar equity or convertible deals.

Are ratepayers protected from data center electricity costs?

Only partly, and mostly where companies volunteer. Microsoft has pledged to cover its own electricity costs and skip local tax breaks, and Anthropic has promised to cover electricity price increases tied to its data centers, part of a nonbinding White House ratepayer pledge signed by tech executives in March 2026. Most states still have no binding rule requiring that outcome.

How much of AI’s capital spending now comes from debt instead of cash flow?

Hyperscaler capital expenditures in 2026 are on pace to consume close to 100 percent of operating cash flow, compared with a 10-year average of about 40 percent, according to estimates from UBS. That gap is what bond and equity markets are being asked to fill.

Disclaimer: This article covers corporate and municipal debt markets for informational purposes only, is not investment advice, and reflects figures accurate as of publication; consult a licensed financial adviser before making investment decisions.