Hyperscaler capex in Q1 2026: the AI build-out gets more expensive, not just bigger

The four largest cloud providers reported first-quarter 2026 results in the last days of April. Cloud growth was strong at all of them. The more useful signal for businesses buying AI services is in the spending lines: how much is going into data centers and chips, how fast it is rising, and what is pushing it up.

The quarter in numbers

The companies do not define capital spending identically, so the table shows each one's own measure and should be read as an approximation rather than a like-for-like ranking.

Amazon's figure is gross purchases of $44.2 billion less $1.0 billion of proceeds and incentives ^[4]. Summed across the four, quarterly spending is on the order of $130 billion, a rough figure given the definitional differences.

Four things worth noting

1. Component pricing is now a named driver. Meta's release says the higher range "reflects our expectations for higher component pricing this year and, to a lesser extent, additional data center costs to support future year capacity" ^[1]. In plain terms, a meaningful part of the increase is that the same hardware costs more, not only that more of it is being bought. The Register likewise reports that Microsoft expects 2026 capital expenditure of about $190 billion, with $25 billion of that due to rising component costs ^[7]. Microsoft's press release itself contains no capex outlook ^[3], so this figure rests on that report of the earnings call.

2. Spending is outrunning cash generation at Amazon. Amazon's trailing-twelve-month free cash flow was $1.2 billion, down 95% from $25.9 billion a year earlier. The company attributed it primarily to a $59.3 billion year-over-year increase in net purchases of property and equipment, an increase it said "primarily reflects investments in artificial intelligence" ^[4].

3. Providers argue demand justifies it. On Amazon's call, CEO Andy Jassy said AWS typically spends "6 to 24 months before we start billing customers," and reported an AWS backlog of $364 billion that excludes a newly announced Anthropic deal he valued at "over $100 billion" ^[6]. Cloud growth supports the case: AWS growth of 28%, Google Cloud at 63%, Azure at 40% ^[2][3][4].

4. Revenue is growing fast but still sits below the outlay. Microsoft's own AI business passed a $37 billion annual revenue run rate, up 123% year over year, according to CEO Satya Nadella ^[3]. That is substantial, yet it is smaller than the roughly $190 billion The Register reports the company expects to spend on capital in 2026 ^[7]. Capital goods also depreciate over several years, so the two figures are not directly comparable, but the gap explains why the question of returns keeps coming up.

Alphabet adds a further wrinkle. Its Q1 capital spending of $35.7 billion was reported alongside Google Cloud revenue of $20.0 billion for the quarter ^[2]. Cloud operating income of $6.6 billion shows the business is profitable at this scale, but the investment is running well ahead of it ^[2].

What it means for pricing

Capacity being built today is the supply that determines token prices and availability over the next two years. Two forces pull in opposite directions: a lot of new supply should keep competition for customers intense, while higher hardware costs and rising depreciation put a floor under what providers can profitably charge. Nothing in these filings predicts a price direction, and we are not making that forecast here. What they do show is that the providers are betting heavily on continued demand, and are exposed if it softens.

For a company buying AI services, the risks are not about the providers running out of money. They are about concentration and lock-in: committing deeply to one provider's models, regions and pricing just as that provider is making its largest-ever infrastructure bets.

Practical implication: Treat AI unit costs as variable rather than reliably falling, build architectures that let you switch models or providers, and review usage-based spend quarterly.