The Largest Capital Sprint in History: Hyperscalers to Invest US$ 630 Billion in AI Infrastructure by 2026
Amazon, Google, Meta, and Microsoft plan to collectively invest up to US$ 630 billion in capital expenditures in 2026, a 62% increase over the record US$ 388 billion in 2025. For technology executives, this concentration of investment reshapes suppliers, pricing, and the geopolitics of global computing.
The capital expenditure figures from the four largest hyperscalers in 2026 redefine what it means to invest in technological infrastructure. Amazon, Google, Meta, and Microsoft plan to collectively invest up to US$ 630 billion in 2026, an approximate 62% increase over the record US$ 388 billion in 2025. For reference, Argentina's GDP in 2025 was approximately US$ 640 billion.
The individual commitments are revealing: Amazon projects US$ 200 billion in CapEx for 2026, compared to US$ 125 billion the previous year. Google announced investments of between US$ 175 billion and US$ 185 billion, more than double the US$ 91 billion from 2025. Meta has committed between US$ 115 billion and US$ 135 billion, up from US$ 72 billion previously. Microsoft anticipates US$ 110 billion to US$ 120 billion, compared to US$ 90 billion.
The New Geography of Computing
This capital is not being invested uniformly. In the American market, 64% of the data centre capacity under construction is located in "frontier markets," areas outside traditional hubs such as Northern Virginia and Silicon Valley. Eight states concentrate over 70% of future capacity: Virginia, Texas, Arizona, New Mexico, Illinois, Nevada, Georgia, and Ohio.
Globally, Google has launched new regions in Sweden, South Africa, and Mexico, with ongoing expansions in Kuwait, Malaysia, and Thailand, as well as US$ 2 billion committed in Turkey over ten years. AWS plans to open a region in Saudi Arabia (US$ 5.3 billion) and the AWS European Sovereign Cloud in Germany (€ 7.8 billion committed until 2040). The Chile region, with over US$ 4 billion, is expected to go live by the end of 2026.
The Energy Bottleneck
The most critical constraint is not capital, hardware, or talent. It is energy. Next-generation AI data centres operate with power densities between 50 and 150 kilowatts per rack, compared to 10 to 20 kilowatts in conventional facilities. Dominion Energy, the primary supplier in Virginia, has issued formal alerts regarding its capacity to meet projected demand.
The response from hyperscalers includes long-term agreements with nuclear power generators, with Microsoft partnering with Constellation Energy (reactivating Three Mile Island) and Google collaborating with Kairos Power for small modular reactors. These decisions have transformed the role of the CFO in AI strategy: 15 to 20-year energy contracts are now integral components of the technological infrastructure roadmap.
What Changes for Enterprise Customers
For enterprises relying on these providers, the capital sprint has concrete implications: increased availability of on-demand GPU capacity at potentially more competitive prices, new points of presence in regions that previously required commitments to smaller local providers, and additional regulatory obligations arising from the designation of critical providers under DORA and equivalent legislation.
The question for the CIO is no longer whether to use a hyperscaler cloud. It is understanding how the concentration of this investment affects their organisation's bargaining power in upcoming service contracts.