- Morgan Stanley projects AI-linked debt issuance will nearly double to roughly $570 billion in 2026 — with about $236 billion already sold by May 31, four times last year’s pace.
- AI bonds are now the largest single sector in the U.S. investment-grade market (about $1.2 trillion by October 2025), surpassing banks — so ordinary 401(k) funds now hold the buildout.
- Apollo and Blackstone are arranging a roughly $36 billion debt deal to buy Google TPUs for Anthropic to lease, with Broadcom backstopping about $31 billion of senior debt.
- Hyperscaler capex is on pace to consume nearly 100% of operating cash flow in 2026 (versus a 40% historical average), so debt — not cash — is now funding the data-center race.
For three years, artificial intelligence was an equity story — record valuations, the largest venture rounds in history, and a sprint to become the most valuable private company on earth. That story is quietly changing. In 2026, the AI buildout is increasingly financed not by selling shares but by selling bonds, and the numbers have grown large enough to touch ordinary retirement accounts. Morgan Stanley projects that AI-linked debt issuance will nearly double to about $570 billion this year. For a business leader, the question is no longer who has the best model — it is who is lending the money, and what happens if the returns arrive late.
From an Equity Story to a Debt Story
How big the borrowing has become
Morgan Stanley estimates that AI-related issuers had already sold close to $236 billion of debt globally by May 31, 2026 — roughly four times the pace of the same stretch a year earlier — on track toward the full-year figure near $570 billion. Over a longer horizon the accumulation is even more striking: by October 2025, AI-linked debt had reached about $1.2 trillion, making it the single largest segment of the U.S. investment-grade credit market and surpassing U.S. banks as the biggest sector in the JPMorgan U.S. Liquid Index, according to M&G Investments. The AI buildout is no longer just a stock-market story; it has become the largest position in the bond market that quietly sits inside index funds and target-date retirement portfolios.
Why hyperscalers suddenly prefer debt
The reason is cash. Hyperscaler capital spending in 2026 is on pace to consume close to 100% of operating cash flow, versus a ten-year average near 40%. Borrowing lets Microsoft, Google, Amazon and the frontier labs keep pouring concrete and buying chips without draining reserves or diluting shareholders through new stock sales. Debt, in effect, has become the pressure-release valve for a capital-expenditure cycle that operating cash flow alone can no longer fund.
Business Insight — For most companies, the lesson is not to copy hyperscaler leverage but to read it as a signal: your AI vendors are funding their roadmaps with borrowed money, which means their pricing must eventually service that debt. Build switching options now, before lenders — not engineers — start setting the price of inference.
The $36 Billion Chip Deal That Explains Everything
An SPV, Google’s chips, and Anthropic’s lease
The clearest window into how this works is a single transaction. Apollo Global Management and Blackstone are arranging a roughly $36 billion debt financing — one of the largest private-credit deals ever — to buy Google’s custom TPU chips. The borrowed money flows through a special-purpose vehicle that purchases the hardware and then leases it to Anthropic for data centers in New York, Texas, Louisiana and Indiana. Anthropic, fresh off a $65 billion raise at a $965 billion valuation, has separately signed more than a dozen U.S. data-center leases exceeding one gigawatt of capacity — a shift from renting cloud capacity toward owned-or-leased compute that improves long-term economics but demands enormous upfront capital.
Who actually absorbs the risk
The structure’s most revealing detail is the backstop. Broadcom, which helps Google build the TPUs, is providing a residual-value support agreement on roughly $31 billion of the senior debt. If Anthropic stops paying its lease and the used chips do not fetch enough on resale to cover the loan, Broadcom absorbs the shortfall. That is vendor financing on a colossal scale — the chip ecosystem effectively guaranteeing the loans that buy its own chips.
Business Insight — This is the circular-financing pattern worth watching. When suppliers underwrite the debt that funds purchases of their own products, demand can look sturdier than it is. It is a powerful accelerant in a bull market and a fragile one the moment utilization or model demand disappoints.
The Grid Becomes the Bottleneck
Washington steps in
Money is only half the constraint; power is the other. On June 18, 2026, the Federal Energy Regulatory Commission issued show-cause orders to all six U.S. regional grid operators outside Texas, directing them to defend or reform interconnection rules so AI data centers can connect to the grid faster without destabilizing reliability or raising consumer bills. FERC Chair Laura Swett called accelerating those connections a “national priority,” bypassing the usual multi-year rulemaking in favor of a targeted, faster path.
The capex behind the urgency
The urgency reflects the spending. Microsoft’s 2026 AI capital expenditure is running near $190 billion; Google has guided to $175–$185 billion; OpenAI’s Stargate joint venture targets 10 gigawatts of capacity. When a regulator fast-tracks grid access for private data centers, it is acknowledging that the AI buildout has become national infrastructure — and, increasingly, it is being financed like it.
Business Insight — Power, not GPUs, may be the real ceiling on AI supply over the next two years. For buyers, that means compute pricing stays volatile and region-dependent; for operators in energy-constrained markets, an electricity strategy is now an AI strategy.
The Question Nobody Wants to Answer
Debt has to be serviced, and that puts the spotlight back on returns. Over the last four quarters, Microsoft’s AI services generated about $37 billion in annualized revenue against roughly $97 billion in spending — a gap equity investors have tolerated but bondholders will scrutinize far more closely. The risk is not that AI fails to pay off, but that the payoff arrives on a slower schedule than the debt. With AI bonds now the largest investment-grade sector, a stretch of disappointing utilization would no longer be a tech-sector problem alone; it would ripple through the index funds, pensions and 401(k) accounts that quietly hold this paper.
Business Insight — Treat AI-vendor stability as a procurement criterion, not an afterthought. Favor providers whose economics work even if capital gets more expensive, keep your data portable, and assume that the cost of AI will, sooner or later, reflect the cost of the debt behind it.
Related
- Oracle Booked $638 Billion. Then Its Stock Crashed.
- Enterprises Bought $401B in GPUs — 95% Sit Idle
- Anthropic Just Made OpenAI the Underdog — Here’s Why
- The Largest IPO Ever Is Secretly an AI Bet
- Nvidia’s Real Moat Isn’t Chips. Qualcomm Just Paid $3.9B.
Sources
- Bloomberg — Apollo, Blackstone Seek Investors for $36 Billion Anthropic Chip Financing Deal
- Yahoo Finance — Morgan Stanley Forecasts AI Debt Issuance to Top $570B in 2026
- VentureBeat — The $401 Billion AI Infrastructure Problem Enterprises Can’t Keep Ignoring
- Build Fast with AI — AI News Today (June 23, 2026): FERC Orders, Anthropic 1GW Leases, AI Debt
AI Biz Insider · AI Business EN · aibizinsider.com
댓글 남기기