Future spending by five tech giants on AI exceeds $3 trillion
If commercial returns fall short of expectations in the next 3–5 years, there will be nowhere to find the money: an analysis
Alphabet, Meta, Microsoft, Amazon, and Oracle have disclosed future obligations for leasing and procurement totaling approximately $3.2 trillion. A significant portion of these is tied to the expansion of AI infrastructure.
Alphabet's obligations amounted to the highest sum—over $1 trillion. This includes equipment, electricity, and leasing; the figure has increased nearly eightfold compared to the previous year.
Meta reported future expenditures of more than $800 billion. The share of data center leasing accounts for slightly less than half: contracts have not yet taken effect, and payments are due over a period of up to 30 years. The total also includes consumer devices from Reality Labs.
The combined volume of future leasing obligations reaches $1.3 trillion, with an additional $1.9 trillion for other procurement. Direct comparisons are complicated by differences in data disclosure; among the accounted items is content licensing. Oracle plans to achieve positive free cash flow by 2028, relying on supplier financing and customer prepayments.
What does this all mean?
1. IT giants are transforming from 'light' software businesses into capital-intensive heavy industry
For decades, Alphabet, Meta, and Microsoft were valued by investors for generating massive free cash flow (FCF) with minimal capital expenditures (CapEx). They did not need to build factories like Boeing or drill wells like ExxonMobil.
Now, due to AI, Big Tech is forced to invest hundreds of billions and trillions of dollars in the physical world: gigawatts of electricity, hectares of land, turbines, concrete, and microchips. In terms of capital commitment, IT giants have become akin to infrastructure monopolies or the heavy industry of the early 20th century.
2. Payback periods have extended beyond the horizon: a bet on 30 years ahead
Increasing Google's obligations eightfold in one year (to $1 trillion) and Meta's to $800 billion with 30-year leasing contracts is an unprecedented risk. In the IT industry, where technologies become obsolete every 3–5 years, signing 30-year contracts for data centers and energy means one thing: companies are binding themselves to obligations that will outlast several generations of hardware and AI models. They are taking on risks comparable to state-level infrastructure projects.
3. Negative cash flow at the most profitable companies in history
The main threat is the emergence of negative free cash flow (FCF) at Oracle, Alphabet, and Meta. This means that even their massive operating incomes (from advertising, search, and cloud services) no longer cover the checks they must write for AI infrastructure. Companies have started dipping into safety cushions and taking on long-term debt just to 'stay in the race.'
4. The risk of an 'infrastructure leverage bubble' emerges
Companies are trying to adapt: Oracle explicitly mentions plans to attract 'customer prepayments and supplier financing.' When infrastructure is built on debt based on future demand that does not yet exist at such volumes, so-called infrastructure leverage is formed. If the commercial returns from generative AI fall short of market expectations in the next 3–5 years, these trillion-dollar obligations for leasing and energy could become an unbearable burden even for those considered 'too big to fail.'
