Hyperscaler

Almost every app you use doesn't just run on your phone. Part of the work happens on other people's computers, which sit in large halls and are accessible over the internet. Such halls are called data centers. A hyperscaler is a company that operates such data centers on an extremely large scale and rents out the computing power within them to other companies. The word combines "hyper" for oversized and "scale" for magnitude. What's meant by this is the ability to double one's own offering within days when demand rises. The largest hyperscalers include Amazon, Microsoft, Google, Meta, and Alibaba.

The Landlords of AI Infrastructure

Anyone who wants to train an AI model today needs thousands of specialized computing chips working simultaneously for months. Buying something like that yourself costs billions. Most AI companies therefore rent instead of building. This puts hyperscalers in a key position: without them, hardly any major AI project can exist.

That also makes them interesting for the stock market. Hyperscalers' investments in new data centers are considered an early indicator for the entire AI industry. When Microsoft and Google increase their spending, the stocks of chip manufacturers like Nvidia usually rise too. If they cut back, the whole industry falls. Analysts call this spending capex, short for capital expenditures.

At the same time, criticism of this concentration of power is growing. Three providers together share about two-thirds of the global cloud market. If one fails, thousands of websites stop working. Such outages have already happened several times, each lasting a few hours.

Why Size Saves Money Here

The economic trick of hyperscalers is called economies of scale. Anyone buying a hundred thousand servers at once pays significantly less per unit than a mid-sized company buying ten servers. The same applies to electricity, cooling, and staff. A single technician in a hyperscaler data center looks after thousands of machines, because almost everything runs automated.

Technically, the offering is based on virtualization. In this process, one large physical machine is divided into many small, isolated computers. So you're not renting a specific computer, but rather a share of a very large pool. If you need ten times as much tomorrow, you're simply allocated more from the pool. This rapid expanding and shrinking is called scaling, and it's the actual core of the business model.

Locations are chosen based on electricity, water, and climate. Cool regions with cheap energy are popular, such as Scandinavia or the northwestern United States. A large data center consumes as much electricity as a small town. That's why hyperscalers now buy wind farms themselves and sign long-term contracts with power plants, sometimes even nuclear power plants.

Where the Term Appears in the News

In everyday life, you encounter hyperscalers constantly without ever reading the name. Netflix streams, Spotify playlists, and the storage of your phone photos run through their data centers. Chatbots like ChatGPT also respond to you from a rented data center. The term itself usually appears in the business section, not in the app.

Typical headlines read: "Hyperscalers raise investments to over 300 billion dollars" or "Hyperscalers build their own AI chips." The latter is an important trend. Google, Amazon, and Microsoft are now developing their own chips to be less dependent on Nvidia.

A common misconception is equating hyperscalers with cloud providers. There are many cloud providers, including small German companies with a single data center. Hyperscalers are only the few really big ones with a worldwide network of locations. Anyone discussing digital sovereignty often means the question of how dependent Europe wants to be on these few providers.