AWS is printing money, but Amazon can’t stop spending it

Amazon dropped its Q1 2026 numbers yesterday, and the headline is familiar: AWS is still the cash cow, but the cost of feeding that cow keeps going up.

AWS revenue hit $28.5 billion for the quarter, beating analyst estimates by a comfortable margin. That’s 22% year-over-year growth, which is higher than I expected given the macro headwinds everyone keeps talking about. The operating margin held steady at around 31%, which tells me the cost optimization wave that plagued AWS through 2024 has finally settled.

But here’s the thing that stood out to me: Amazon’s total capital expenditures hit $18.2 billion in Q1. That’s up 35% from last year, and CEO Andy Jassy made it clear during the earnings call that this isn’t a temporary spike. He said the company expects CapEx to increase in the coming quarters, with most of it going toward AWS infrastructure — data centers, networking gear, and the custom AI chips they’ve been developing.

This is the same pattern we’re seeing across the entire cloud industry. Microsoft and Google are throwing money at AI infrastructure too, but Amazon’s spend is particularly eye-catching because they’re already the market leader. You’d think they could coast a bit, but nope — Jassy is doubling down.

What’s driving this? AI workloads, obviously. The demand for GPU instances and Amazon’s own Trainium chips is outpacing supply, and Jassy said they’re “building capacity as fast as we can.” That’s a polite way of saying they’re in an arms race with Microsoft Azure and Google Cloud, and nobody wants to be the one caught short when the next wave of AI startups starts scaling.

The interesting part is how investors are reacting. The stock dipped slightly after hours, which I think is a knee-jerk reaction to the CapEx number. But honestly, if AWS is generating $9.5 billion in operating income per quarter, they can afford to spend. The question is whether those investments will pay off before the next downturn hits.

I’ve been around long enough to remember the last cloud spending spree in 2020-2021. Everyone was building data centers like there was no tomorrow, and then growth slowed and margins got squeezed. This time feels different because AI demand is real and sustained, but the risk is still there. If the AI hype cycle cools off faster than expected, Amazon could be sitting on a lot of underutilized hardware.

For now, though, Jassy is betting that the AI infrastructure buildout will be as transformative as AWS itself was a decade ago. He might be right. But I’d be watching those utilization numbers closely in the next few quarters.