Hyperscalers bet big on infrastructure, returns still unclear
Investments in the construction and modernization of data centers are now reaching a level comparable to the economies of entire states. And not the small countries are meant. Amazon alone has spent more than $100 billion on infrastructure development over the past year. This is about the same amount as the entire economy of Costa Rica produces in a year, and more than the GDP of Luxembourg or Lithuania.
Other cloud market leaders have a similar picture:
- Google has spent $82 billion on this, which is more than the entire annual volume of the Slovenian economy.
- Microsoft’s figure is slightly lower, $75 billion, which is still higher than Uganda’s GDP.
- Meta has invested $69 billion in its data centers, which is more than Bahrain’s annual production.
If we look at the industry as a whole, spending continues to grow rapidly. Capital expenditures on data centers are projected to exceed $657 billion in 2025. For comparison, in 2023 it was about $330 billion. The growth has almost doubled in two years, primarily due to the race for AI capabilities. The question is whether today’s record investments will deliver quick returns, or whether the industry will continue to work for the future. Currency markets are already responding, with USD JPY performance often serving as a barometer of investor confidence in these long-term tech bets.
Hyperscalers expect these investments to bring income someday, but for now, it’s more of a long-term bet. Even players like xAI, which are not among the largest providers, spend about a billion dollars a month on infrastructure.
However, not everyone is confident that large-scale investments will pay off quickly. In the second quarter of 2025, Meta explicitly stated that the time-tested machine learning models underlying its recommendation systems were generating profits. Generative AI still remains in the category of promising, but not among the primary revenue streams. Nevertheless, cloud companies are ready to offer more and more AI resources to those who want to develop their own projects.
There may be a shortage of computing power in the market in the coming years. New models, such as the recently introduced GPT-5, require more and more resources. At the same time, the need for infrastructure for inference, the daily use of AI is also growing. Already today, over 700 million people use ChatGPT alone, of which over 120 million use it daily.
Although IT equipment will remain the main focus of industry spending, investments in physical infrastructure will grow fastest. We are talking about power supply systems, energy distribution, and cooling. The computing density in racks is growing almost exponentially, which demands ongoing technical upgrades.
At the same time, there is an increasing interest in local generating capacities and microgrids. Major operators are already investing in these areas to reduce their dependence on external power sources. It is also expected that the construction of data centers will accelerate, and the largest sites will reach a capacity of at least one gigawatt.
Some projects are already beyond their usual scope. Meta has announced several campuses with a total capacity of several gigawatts, which are planned to launch in 2026. Individual sites at an early stage of design may surpass 5 GW.
All this means that the data center market is changing rapidly. The question is whether today’s record investments will be able to bring a quick return, or whether the industry will continue to work for the future, increasing costs and pushing infrastructure to the limits of its capabilities.