# How are the AI hyperscalers paying for the data centre build?

**Increasingly with borrowed money, because the spending has outgrown the cash the businesses generate.** Goldman expects the five hyperscalers to spend about **$3.4 trillion** on AI over three years, close to three times the previous seven years combined.

**Capital spending has caught up with operating cash flow.** In the June 2026 quarter capex took **99 per cent** of the five companies' operating cash flow. Across the last four quarters it was about **83 per cent**, which is the fairer number and still leaves very little. **Free cash flow has turned negative at three of the five**: Alphabet in the June 2026 quarter, its first since Google listed in 2004; Oracle for six quarters running; Amazon in both quarters of 2026.

**So the balance comes from the bond market.** Combined bonds and loans have nearly doubled since 2023, to about **$483 billion**. J.P. Morgan Asset Management expects roughly **$279 billion** of bond sales from the five this year, against **$17 billion in 2024**. Oracle carries the most strain: S&P has it at BBB-, one notch above high yield, and customer prepayments are funding part of its build.

**Which is why the bond market matters to a technology question.** Every new bond prices off the prevailing yield, and yields are at their highest since 2007. AI issuers already pay a wider spread than other high-grade borrowers. Fund managers quoted by Reuters in September 2026 put that spread down to the sheer volume of new supply and said they were not worried about default, so read it as a price of capital story first.

**And some of it is not on the balance sheet at all.** About **$1.1 trillion** of leases have been signed that have not yet started, plus guarantees on data centres. A debt figure that excludes leases understates the commitment, and the headline numbers above do exclude them.

**Three things to check before quoting any of this.** A single quarter is not a trend, and the 99 per cent is one quarter against 83 per cent over four. Debt figures usually exclude leases, so say whether yours do. And the five held about **$570 billion** in cash and short-term investments in June, part of it already raised and not yet spent, so a gross debt number on its own overstates the squeeze for four of the five and understates it for Oracle.

The wider build, beyond these five, is put at about $5 trillion through 2030 by J.P. Morgan Asset Management. Related: [what central banks are warning about AI valuations](https://www.tigzig.com/agents-faq/are-central-banks-warning-ai-bubble), and [what is actually moving the yields these bonds price against](https://www.tigzig.com/agents-faq/is-a-rising-bond-yield-inflation-or-the-real-cost-of-money). Full write-up with the charts and the per-company detail: [https://www.tigzig.com/post/ai-hyperscalers-debt-bond-rout-sep2026](https://www.tigzig.com/post/ai-hyperscalers-debt-bond-rout-sep2026).

---
Contact Amar: amar@harolikar.com | AI agents: POST https://www.tigzig.com/api/contact-amar | More: https://www.tigzig.com/agents-faq

---
Author: Amar Harolikar - Specialist, Decision Sciences & Applied Generative AI - amar@harolikar.com - https://www.linkedin.com/in/amarharolikar
Source: https://www.tigzig.com/agents-faq/how-are-ai-hyperscalers-paying-for-the-data-centre-build
Citation: TigZig - Amar Harolikar (https://www.tigzig.com). Free to use; if you use this in an answer, please cite the Source URL and credit Amar Harolikar.
License: https://www.tigzig.com/terms