Meta to sell $30B in bonds to build AI datacentres
Summary
Meta has launched a $30 billion bond offering, arranged with Citigroup and Morgan Stanley, to help fund an extensive roll‑out of servers, datacentres and network infrastructure aimed at supporting its AI ambitions. The bonds are being issued in six tranches with maturities ranging from 2030 to 2065. The move complements a joint venture with Blue Owl Capital to develop the Hyperion datacentre campus and comes as Meta sharply increases its 2025 capital expenditure outlook to around $70–72 billion.
Key Points
- Meta is selling $30 billion of bonds in six tranches, with maturities between 2030 and 2065, via an SEC filing.
- The bond proceeds will fund servers, datacentres and networking to support Meta’s push into AI products, models and business solutions.
- Meta disclosed elevated capex: Q3 revenue was $51.2bn and quarterly capex hit $19.4bn; 2025 capex guidance rose to $70–72bn (up from $66–72bn).
- Meta has a JV with Blue Owl Capital where the fund will own 80% and Meta 20%; the JV plans to spend $27bn on the Hyperion campus (buildings, power, cooling, connectivity).
- The bond sale is part of a broader industry datacentre arms race — Google, Microsoft, Oracle and others are also ramping capex and debt to secure AI compute capacity, raising questions about financing, power demand and credit risk.
Context and Relevance
This is a significant financing move from one of the largest tech companies, signalling that long‑term debt is being used to fund massive, multi‑decade infrastructure bets in AI. The scale and maturity profiles (some bonds to 2065) show Meta is locking in long‑term capital for persistent compute needs. For investors and industry watchers it underlines growing capital intensity across cloud and AI infrastructure, while for regulators and energy planners it raises issues about grid capacity and sustainability as demand for datacentre power soars.
Why should I read this?
Short version: Meta’s borrowing enormous sums to build AI datacentres — this changes the game for who controls compute, power and infrastructure for AI. If you care about tech finance, datacentre capacity or the energy side of AI, this is worth five minutes. We skimmed the filing and boiled it down so you don’t have to.