Tech Companies Have Issued $244B in Bonds in 2026 to Finance AI Infrastructure: It's a Lot

Yes, this is a lot. The $244B number represents just Amazon, Alphabet, Microsoft, Meta, and Oracle. This is about twice the 2025 number, and we are only in July. Why?

Why do hyperscalers need money? It feels like there is insatiable demand for inference or tokens (i.e., massive computing power). At its I/O event this spring, Google reported that a few large clients like Uber had blown past their annual token budget in just the first quarter. The cost of building data centers (and training models) is increasing due to inflation caused by supply shortages (e.g., electricity, construction, semiconductor chips, cooling systems, etc.). Apple's Tim Cook referred to the memory price surge as an "unprecedented challenge." The bill of materials for an iPhone that relies on similar components is up $300.

Why bonds? For starters, spending is outpacing cash. Just two to three years ago, a typical hyperscaler's free cash flow (FCF) would be more than two times its capital expenditure. Now that equation has flipped. Not only is capex growing, but initial 2026 estimates made in the winter were already inaccurate by spring. Tech companies have good credit, so the cost of borrowing has been low. Infrastructure is a different business model for most. Duration-matching (of the asset) logic is at work. Also, issuing bonds allows them to hang on to cash for acquisitions. There are more financial reasons. For me, the other factor is that it broadens the audience for AI investments to institutional investors that must maintain a balanced portfolio. (Besides, how many circular deals can one do, i.e., "you invest in me and I will funnel the money back to you.")

Is the strategy working? Mostly, but the market's energy is waning. Meta's $30B bond offering in October 2025 attracted $125B in investor demand. Now, the market's ability to absorb these bonds is shrinking. When that happens, the rate — or cost of debt — increases for them. On one hand, there is a lot of supply. On the other, the risks keep mounting. Risks include: over-rotation from humans to AI agents (i.e., AI agents don't necessarily save money), capex spend outpacing revenue, communities fighting new data centers, regulation lurking, lower-cost models from China (e.g., Kimi) threatening existing revenue, and courts starting to hold tech companies liable for what happens on their platforms. This isn't the entire bond story, but it will get you started.