Seven companies—Google, Meta, Amazon, Microsoft, Oracle, Nvidia and Broadcom—have committed to spending about $3 trillion on AI infrastructure outside their reported budgets, according to a report examining the sector’s commitments. The spending is concentrated mainly in data-center leases and the purchase of chips, memory modules and networking equipment.
Leases and equipment account for most of the spending
Of the commitments, $1.1 trillion is earmarked for leasing data centers that have not yet begun operations, as lease agreements take effect before construction is completed. The second component consists of about $1.7 trillion allocated to buying chips, memory modules and networking equipment.
These figures do not include the cost of cooling processors or the electricity needed to run them, as well as the labor and technical expertise required to operate data centers.
Anthropic expects to spend $518 billion on cloud computing over the next decade, while OpenAI expects to spend about $860 billion through 2030. These commitments come alongside plans to float shares in the two companies, at valuations of up to $2 trillion for Anthropic and $1.4 trillion for OpenAI, according to reports on the plans.
Anthropic’s valuation is based on earnings projections and AI development costs, with earnings expected to exceed $100 billion by the end of the current year.
Servers top the budgets of major companies
Alphabet Chief Financial Officer Anat Ashkenazi said during the company’s second-quarter 2026 earnings call that its projected AI spending had reached $190 billion for 2026. The company allocated 60% of its infrastructure spending to servers and 40% to data-center buildings and networking equipment.
Microsoft Chief Financial Officer Amy Hood said during the company’s fourth-quarter fiscal 2026 earnings call that about two-thirds of its spending went toward short-lived assets, referring to the processors and graphics cards needed to run AI. The company’s spending on this component reached $41 billion in the fourth quarter of the fiscal year.
The short operating life of these assets means that upgrading data centers requires the purchase of new processors and chips to meet growing computing needs, as component prices rise amid a memory and processor shortage.
Amazon raised its 2026 cash-spending forecast to about $220 billion, according to comments by Chief Executive Andy Jassy during the company’s second-quarter 2026 earnings call with investors and shareholders.
Demand exceeds data-center capacity
Companies face a time gap between the start of spending on data centers and the point at which they become operational and generate revenue, while demand for cloud-computing capacity continues to exceed available supply.
Jassy said Amazon had recorded strong demand for its data centers through 2028, although it would not be able to meet all demand in 2026 and 2027. Ashkenazi, meanwhile, said the backlog of orders for Google’s cloud services had reached $514 billion.
Hood said demand from Microsoft customers continued to exceed available supply, prompting the company to double its production capacity over the next two years.
Constraints on electricity supply and consumption add another challenge to the expansion of data centers, as growing demand requires the construction of new facilities and the securing of additional energy sources. As companies increasingly adopt AI technologies, pressure is mounting on cloud-computing capacity as a limited resource facing growing demand.
