Hyperscaler bond tracker

How to read the hyperscaler bond tracker

Building data centers for AI costs more than even the largest technology companies generate in cash, and a growing share of the gap is being borrowed. This page follows that borrowing for the five hyperscalers — Amazon, Alphabet, Meta, Oracle and Microsoft — deal by deal, and puts it next to the one part of the bond market most people feel directly: mortgage rates.

Where the data comes from

Registered deals come from the pricing term sheet each issuer files with the SEC on the day a bond prices. It lists every tranche — a single deal is often six to eleven bonds of different maturities — with its size, coupon, maturity, re-offer price, yield and spread over the benchmark government bond. Off-balance-sheet deals are private 144A placements that file no term sheet. They are identified from the project company that issued them, matched to their bond listings, and sized from the sponsor’s pricing announcement.

Off-balance-sheet debt

Not all of the borrowing shows up as a hyperscaler bond. Increasingly a separate project company borrows to build a data center that a hyperscaler will use, and the bonds are repaid from that use. The hyperscaler’s link can be a lease (it is the tenant), a lease backstop (it guarantees a smaller tenant’s rent), or a joint venture (it co-owns the project and guarantees its value). The largest example is the $27.3 billion Beignet Investor bond that financed Meta’s Hyperion data center in Louisiana — larger than most of Meta’s own bond deals, and not on Meta’s balance sheet.

How big is it?

Against the whole U.S. bond market, hyperscaler borrowing is small: a couple of percent of all bond issuance, and under one percent of Treasury sales once short-term bills are counted. Against the corporate market it is large for a handful of companies — a high-single-digit share of all corporate bonds sold. Where it matters most is the long end: measured against the 20- and 30-year bonds the Treasury itself sells, hyperscaler bonds maturing fifteen or more years out are a meaningful fraction, because that is where they compete with the Treasury and with mortgage investors for the same insurers and pension funds.

Reading the spread

A bond’s new-issue spread is the extra yield it pays over a Treasury of similar maturity at the moment it is sold, in basis points (hundredths of a percentage point). It is the cleanest measure of what investors demand to lend to a company, because it strips out the general level of interest rates. When the same issuer has to pay a wider spread for the same maturity deal after deal, investors are asking more to absorb more of its debt. The long end matters most here: hyperscalers have been selling unusually long bonds, 30 and even 40 years, which compete directly with other long-duration assets.

Mortgages and crowding out

Mortgage rates are priced off mortgage-backed securities, which compete for the same long-duration fixed-income investors — insurers, pension funds, bond funds — as high-grade corporate bonds. A heavy supply of long-dated, highly rated corporate debt can widen the spread those investors demand across the board. The crowding-out chart plots the 30-year mortgage rate’s spread over the 10-year Treasury against hyperscaler issuance so the relationship can be judged directly. Many other things move mortgage spreads — rate volatility, bank demand, the Federal Reserve’s own mortgage holdings — so a correlation on this chart is context, not proof of cause.