What it is. A hyperscaler is a cloud provider operating at a scale where its purchasing and product decisions reshape the markets it touches. The unofficial canon: Amazon Web Services, Microsoft Azure, and Google Cloud Platform. Oracle Cloud Infrastructure gets included now thanks to its AI-training book. Meta is usually included on capex discussions even though it doesn't sell external compute, because its data-center build matches the others in scale. Some analysts add Alibaba Cloud for the Asia-Pacific picture.
Why the label matters. Hyperscalers are the buyers that move the AI hardware market. When Microsoft commits to a multi-year Nvidia GPU contract, that single decision can swing Nvidia's quarter and the chip industry's outlook for the next 18 months. When the four U.S. hyperscalers collectively guide capex up by $50 billion in a quarter, semiconductor equipment makers (ASML, Lam Research, Applied Materials) reset their order books accordingly. The cascade goes further: power utilities in Virginia, Ohio, and Texas plan multi-year grid investments around hyperscaler data-center load growth.
The capex print is the number to watch. Each hyperscaler reports forward capex guidance with earnings. Stack them up and you get the leading indicator for the entire AI build-out. The combined Microsoft + Google + Amazon + Meta capex line is now well into the hundreds of billions annually and growing. That's the number the AMD and Nvidia stocks trade on, and increasingly the number the regional power markets price off.
What's changing. The traditional cloud-services revenue line at the hyperscalers is now competing for capex with the AI training and inference build. Some quarters that creates margin pressure (cost of revenue rising faster than top line); some quarters the AI book pulls margins up via premium pricing on H100/B200 instances. The mix shift is the story embedded in most of our Big Tech earnings coverage.
Who is not a hyperscaler. CoreWeave, Lambda, and the newer GPU-cloud specialists run at hyperscale-adjacent capex but lack the diversified services base. They are usually called "neoclouds" or "GPU clouds" to distinguish them from the original four. Snowflake, Databricks, and Salesforce build on top of hyperscalers, not alongside them.
Bottom Line
"Hyperscaler" is shorthand for the four-to-six companies whose purchasing decisions are the closest thing the global tech economy has to a single demand signal. When you see "hyperscaler capex," read it as the AI build-out's headline number.