Investors are observing the amount that Big Tech spends on artificial intelligence each quarter, the more significant figure being the amount that has already been locked in by contract for years to come.
For much of the artificial intelligence boom, Wall Street has kept a close eye on the capital expenditures of Microsoft, Meta, Amazon, and Alphabet. Although these figures are huge, they do not now convey the full picture. By June 30, the four Magnificent Seven companies had made public around $830.5 billion in future payments relating to leases which had not yet started. Meta then revealed an additional approximately $68 billion from data-centre leases entered into in July, bringing the total known amount to about $898.5 billion. Amazon’s figure is not quite comparable since its lease portfolio also covers warehouses, offices, aircraft and vehicles, but the scale is still hard to ignore.
These commitments are not concealed debt; they are stated in the companies’ official filings, usually involve undiscounted payments spread over a number of years, and should not be treated as simply another form of borrowing. It is precisely this distinction that gives them their importance. A large part of the infrastructure being ordered in connection with the AI boom has already been economically committed before it shows up as a conventional lease liability on the balance sheet. For investors who assess the Magnificent Seven based on earnings growth, free cash flow, and return on invested capital, the amount of capital expenditure reported might be only the starting point of the calculation.
AI Is Quietly Moving Wall Street’s Buyback Machine Downstream →
Microsoft is perhaps the best example of why reported capital expenditure can become misleading. As stated in Microsoft’s fiscal 2026 fourth-quarter earnings call, Microsoft is increasing the estimated useful lives of its data centres and office buildings from 15 years to 25 years starting in fiscal 2027. This accounting change will result in a greater number of future data-centre leases being categorised as operating leases rather than finance leases, since finance leases are part of Microsoft’s capex figure while operating leases are not. What is the result? Microsoft has reduced its expected capital expenditure for calendar year 2026 to about $175 billion, even though it has made it clear that its original investment expectations have not changed.
The physical AI build-up wasn’t necessarily reduced in any way; instead, the way in which part of that investment was presented changed. That is important since the information in Microsoft’s latest Form 10-K shows that there are $329.1 billion worth of further leases, mostly relating to data centres, which had not yet started as of June 30. The leases are expected to start between fiscal year 2027 and fiscal year 2033 and may last as long as 20 years.
There is an even greater difference in Meta’s cash flow. In the second quarter the company spent $31.08 billion on capital expenditures, comprising the finance-lease principal payments, and managed to generate only $784 million of free cash flow. Meta now expects its capital expenditures to amount to approximately $130 billion to $145 billion in 2026. At the same time, its June filing revealed that there were approximately $278.99 billion of leases that had not yet commenced, together with another $68 billion of data-centre leases signed in July. The problem isn’t that investors are unable to see the spending; it’s that a well-known capex figure can make a considerably larger future fixed-cost structure appear more manageable than it actually is.
Bank of England Scrutinizes the Debt Behind Britain’s AI Investment Boom →
The more significant change could be taking place in the way the business is structured. In the past, large technology companies have enjoyed high valuation ratings to some extent since huge revenues could be generated without needing a corresponding increase in physical assets. This situation is now being changed by AI. Shay Boloor, who is the chief market strategist at Futurum Equities, has recently said to Reuters, “Investors are underestimating how fundamentally AI is changing the Big Tech business model.”
The figures support that argument. Alphabet revealed that it had $85.2 billion worth of leases, mostly concerning data centres, which had not yet commenced by June 30. In the first six months of 2026 the company spent $80.6 billion on capital expenditures. Amazon stated that it had about $137.2 billion of leases that had not yet started and spent $96.3 billion on cash capital expenditures during the first six months of the year. Alphabet’s latest filing details the scale of its own infrastructure commitments.
Even Nvidia, although it is receiving a large part of this spending, is becoming more committed in terms of capital. The most recent document it filed shows $366 billion in future commitments, comprising $279 billion in commitments relating to supply and capacity, $29 billion in agreements for cloud services, $25 billion in data-centre leases which have not yet started, $25 billion in planned equity investments and $8 billion in capital expenditure commitments. The various categories are distinct and should not be treated as debt, yet they show how deeply future AI growth is already being incorporated throughout the supply chain.
The question changes for finance investors. The right metric might now go beyond just capex. Investors are becoming more inclined to take into account ‘capex plus future contractual infrastructure exposure’ and then compare that commitment to the future incremental operating cash flow. An analysis by Reuters of estimates from LSEG shows that Microsoft, Alphabet, Amazon, Meta and Oracle are on course to spend more on combined capital expenditure than the amount of free cash flow they generate by 2027. During the period from 2025 to 2027, expected annual growth in operating cash flow of about $340 billion will be matched by approximately $534 billion in extra capital expenditure, which amounts to roughly $1.57 of additional investment for each $1 of extra operating cash flow. It is at this point that AI ceases to be just a growth story and turns into a capital-efficiency story.
Five Beaten-Down Stocks Still Riding the AI Boom →
One strong counterargument exists. The companies known as the Magnificent Seven are not ones that are worried about whether customers will turn up and so sign lease agreements. Microsoft, Amazon, Alphabet and Meta run some of the most profitable technology platforms in history and there is still a great deal of demand for cloud and AI computing. If the rate at which AI is consumed continues to rise, it might turn out to be the correct approach to secure land, electricity, chips and data-centre capacity many years in advance. The limited capacity obtained today could then support huge revenue streams in the future.
The hundreds of billions of dollars involved should not be regarded as something that will have to be paid right away; these commitments extend over years or decades, many of the facilities have not yet opened, and the way they are accounted for is very different from that of ordinary corporate debt. Even before they formally appear as lease liabilities, Reuters has pointed out that the rating agencies can include these obligations in their adjusted leverage calculations. That is the reason why the bearish argument shouldn’t be that Big Tech has secretly accumulated nearly $900 billion in debt since it has not.
What is more interesting is that the businesses of the Magnificent Seven which have been most exposed to AI are at the same time becoming more capital intensive while investors expect artificial intelligence to make them more productive. The tension will ultimately have to be settled in the cash-flow statement. Microsoft, Meta, Amazon and Alphabet need their AI-related revenue and operating cash flow to grow quickly enough so that the infrastructure investments made today yield attractive returns instead of merely resulting in higher depreciation, lease payments and financing needs.
Investors ought therefore to pay more attention to lease commencements, operating lease expenses, depreciation growth, free-cash-flow conversion, cloud revenue, incremental margins and return on invested capital since these factors may show more clearly whether the AI buildout is in fact creating economic value. The problem with the Magnificent Seven does not require AI to fail in order to arise. Even if AI is revolutionary it can give rise to unsatisfactory investment returns if a large amount of capital is committed too early.
Wall Street is already aware of how much money these companies are spending today. The next question regarding valuation is what will occur when the bills which they have already signed start to arrive.
[…] Wall Street Is Watching AI Capex. The Real Bill Is Still Coming → […]