Hyperscalers raised $108B debt in 2025

The story here is corporate debt — but really it’s about AI infrastructure. Training and serving frontier models now means buying land, power, networking gear, GPUs, and whole campuses of data-center capacity years before the revenue is fully visible. Big tech used to fund most of that from cash flow. Now even the richest companies are reaching for bond markets because the buildout got too large to treat as a normal capex cycle. Amazon, Meta, Oracle, and Alphabet have all either sold big chunks of debt or laid out fresh financing plans tied to AI expansion over the past several months. (newsbreak.com) ### Why borrow if these companies are already rich? Because AI infrastructure is lumpy. A company can be wildly profitable and still not want to drain cash reserves into one giant hardware-and-power sprint. Debt lets management spread the cost over time, keep buybacks or acquisiti(newsbreak.com)id it was investing $80 billion in AI-enabled datacenters, while Azure revenue topped $75 billion in fiscal 2025. (datacenters.microsoft.com) ### Which companies actually raised the money? Amazon raised $15 billion in a U.S. dollar bond sale in November 2025 — its first such sale in three years — with Reuters noting the proceeds could go to acquisitions, capex, and buybacks as AI infrastructure spending ramped. Meta then sold $25 billion of investment-grade bonds on April 30, 2026 after lifting its AI spen(datacenters.microsoft.com)et in May 2026 after already raising about $32 billion across dollar, sterling, and Swiss franc markets. (newsbreak.com) ### Why is Oracle in the middle of this? Because Oracle is trying to jump from “important cloud vendor” to “must-have AI capacity partner.” In February 2026 it said outright that it expects to raise $45 billion to $50 billion during calendar 2026 to expand Oracle Cloud Infrastruct(newsbreak.com)ity issuance. That is unusually explicit — and it tells you Oracle sees contracted demand big enough to justify financing first and building second. (oracle.com) ### How big is the underlying demand? Big enough that Oracle’s numbers stopped looking like normal software-company numbers. In June 2025, Oracle said its remaining performance obligations hit $138 billion, and management said cloud infrastructure growth should accelerate from 50% in fiscal 2025 to above 70% in fisca(oracle.com)ert to near-term revenue, but it does show customers are locking in future capacity at huge scale. (investor.oracle.com) ### Is this only happening on company balance sheets? No — the financing chain is spreading outward. In April 2026, Blackstone-backed QTS sold $4.6 billion of bonds tied to a data-center project in Georgia that Bloomberg said would house thousands of ser(investor.oracle.com)me AI demand wave. (bloomberg.com) ### So where does the “$108 billion” idea come from? Basically from adding up a cluster of jumbo deals and planned financings across the hyperscaler ecosystem. But the catch is timing. The clean, well-sourced picture is not “hyperscalers raised exactly $108 billion in 2025.” It’s that a debt-financed AI b(bloomberg.com)ping capital markets in size. (newsbreak.com) ### What’s the real takeaway? AI infrastructure has crossed a line. It is no longer just a capex line item inside giant tech companies. It is now a financing regime — one where access to debt, equity, power, and supply chain all matter as much as model quality. The winners may not be the companies with the best demo. They may be the ones that can fund the next three years of compute first.