Morgan Stanley says AI debt issuance will more than double to nearly $570 billion this year. Big Tech's cash machines can no longer fund the buildout alone. The bond market is now the most important component in the silicon supply chain.
By Blumefield | June 11, 2026
A Forecast That Redraws the Map
Morgan Stanley dropped a number on Wednesday that reframes how the AI boom actually works. The bank now expects global AI debt issuance to more than double this year, reaching nearly $570 billion in 2026. By the end of May, AI-related issuers had already sold roughly $236 billion of bonds worldwide, about four times what they raised over the same stretch of 2025. If the forecast holds, the financing of artificial intelligence will rank among the largest corporate borrowing waves in history, bigger than the telecom buildout of the late 1990s and rivalling the leveraged buyout booms that preceded past credit cycles.
The headline figure matters less than what sits underneath it. For two decades, the giants of American technology were famous for needing nobody's money. Alphabet, Microsoft, Amazon and Meta generated so much cash that their occasional bond sales were treated as balance sheet housekeeping. That era is ending in real time. Morgan Stanley estimates the four hyperscalers will spend roughly $700 billion on AI-related capital projects this year, a figure the bank expects to push past $1 trillion in 2027.
Spending of that scale has outgrown even the most prolific cash machines ever built. The result is a structural shift: the AI race is no longer funded by profits alone. It is funded by lenders.
When Cash Flow Stops Being Enough
The arithmetic behind the borrowing surge is stark. Hyperscaler capital expenditure in 2026 is on pace to consume close to 100 percent of operating cash flow, against a ten-year average of around 40 percent. Companies that once paid for data centres out of spare change now face annual infrastructure budgets larger than the cash their businesses produce. Selling new equity would dilute shareholders at a moment when boards are under pressure to show AI returns, so AI debt in the form of investment-grade paper has become the path of least resistance.
Morgan Stanley expects net bond supply from the hyperscalers alone to rise between 30 and 50 percent this year, landing somewhere between $130 billion and $150 billion. The firms are also casting a wider net for buyers. The bank noted that hyperscalers have been broadening their investor base through non-dollar issuance, tapping euro and other currency markets to avoid saturating any single pool of demand.
That detail deserves more attention than it has received. When the world's most creditworthy borrowers start engineering their issuance to avoid exhausting investor appetite, it signals that the sums involved are beginning to strain even the deepest capital markets on earth.
The $1.5 Trillion Hole
Behind the bond totals sits a larger and less visible funding problem. Morgan Stanley's credit research, laid out in the bank's Thoughts on the Market analysis, identifies a financing gap of roughly $1.5 trillion that must be met by external capital through the buildout. Around $800 billion of that is expected to come from private credit, with a further $200 billion from investment-grade bonds issued by the large technology names, and the remainder from a patchwork of securitisations, joint ventures and project-level structures.
Private credit's role is the part regulators will be watching. Bond markets are transparent, priced daily and stress-tested by decades of cycles. Private lending to data centre developers, GPU leasing vehicles and power projects is none of those things. As Blumefield has reported in its coverage of AI compute financialisation, Wall Street has already begun wrapping chips and compute contracts into tradeable instruments. The AI debt complex now stretches from blue-chip bonds at one end to bespoke private structures at the other, and visibility worsens at every step down that chain.
The chipmakers add another layer. Morgan Stanley notes that semiconductor financing is accelerating across public and private markets, with deals increasingly structured as shorter-term instruments that amortise fully over their life. That structure tells its own story: lenders want their money back before the next chip generation makes today's collateral obsolete.
What Bond Investors Are Really Buying
For the people whose savings sit in index funds and target-date retirement accounts, this shift is not abstract. The AI buildout is becoming one of the largest single exposures in the investment-grade credit market, which means ordinary pension and 401(k) portfolios are now lenders to the data centre boom whether their owners know it or not.
For now, the credit market is sanguine. Morgan Stanley observed that fundamentals remain strong and that bond price action is being driven mostly by supply expectations rather than worries about repayment. Spreads on hyperscaler AI debt remain tight by historical standards. These are, after all, companies with fortress balance sheets, vast recurring revenues and credit ratings that rival sovereigns.
But the history of great infrastructure buildouts suggests the risk rarely lives where everyone is looking. The railways were transformative and still ruined their bondholders. The fibre laid in the dot-com era carried the internet for decades after the firms that financed it went bankrupt. The technology succeeding and the debt performing are two different questions, and the gap between them is where credit cycles are made.
The Discipline Debt Brings
There is a case that this is healthy. Equity tolerates dreams; debt demands schedules. A hyperscaler that funds AI expansion from its own cash answers only to a patient shareholder base. One that borrows $50 billion answers to covenants, coupons and refinancing windows. As AI debt becomes the industry's primary fuel, capital discipline arrives with it, and projects that cannot demonstrate a credible path to cash generation will find the next tranche harder to raise.
That discipline cuts both ways. Debt-funded buildouts cannot simply pause when sentiment cools. Interest must be paid in downturns as well as booms, which converts the AI capex cycle from a discretionary spending programme into a fixed obligation woven through the global financial system. The Federal Reserve and other supervisors have spent two years insisting that AI exuberance was an equity phenomenon, contained and loss-absorbing. With more than half a trillion dollars in AI debt set to price in a single year, that assessment is now out of date.
The first chapter of the AI era was written in research labs. The second was written in data centres. The third, it is now clear, will be written in the bond market, and it is the one chapter where the ending is never optional. Blumefield will track every issue as it prices. The buildout has found its banker; what it needs next is a payback period.