Wall Street is starting to treat Big Tech’s AI build‑out as infrastructure debt: trillions in on‑balance and hidden commitments that only make sense if AI revenues arrive fast enough to cover swelling data‑center and accelerator costs.
AI infrastructure is turning high-margin software giants into capital-intensive utilities—and the debt behind that shift is only just starting to show up in the numbers.
The artificial intelligence boom has been the main driver of growth in the technology sector for several years, but it remains unclear how sustainable the current model is, with the world’s largest technology companies continuing to increase spending on computing infrastructure. The fact is that financial reporting does not always reflect the real scale of financial obligations undertaken by the industry.
The combined debt of Alphabet, Microsoft, Amazon, Meta, and Oracle has already reached $1.65 trillion, adding to investor caution that has weighed on Dow Jones futures. At the same time, about $1.35 trillion is reflected in their financial statements, while another $300 billion consists of various forms of off-balance-sheet commitments, primarily related to the construction of data centers for artificial intelligence.
This is a standard accounting practice. A significant portion of these obligations arises under long-term agreements to build or lease infrastructure and is not reflected in the companies’ balance sheets before the facilities become operational. However, once construction is completed, these costs will inevitably end up in the financial statements, increasing the debt burden of the largest market players.
The trend is particularly evident at Meta. The volume of its off-balance sheet liabilities reaches $420 billion, nearly three times more than the company’s officially recorded debt. Oracle has increased similar commitments by more than 30 times over the past four years to $273 billion. This is largely due to the company’s participation in the large-scale Stargate project, which involves the construction of new data centers for the development of AI infrastructure in the United States.
All this comes against the backdrop of an unprecedented increase in the industry’s capital expenditures. This year alone, the largest American hyperscalers plan to spend up to $725 billion on building data centers and purchasing AI accelerators. The spending is being funded through a combination of operating cash flow, bond issuance, and structured financing arrangements.
Over the past five years alone, Alphabet, Amazon, Meta, Microsoft, and Oracle have increased their debt burden by about $350 billion. Although their debt servicing costs remain manageable due to the high profitability of their businesses, investors are paying increasing attention to the returns generated by these investments.
Amazon’s recent $25 billion bond issue was met with a noticeably cooler market reception than similar offerings in previous years. For many market participants, this was the first sign that investors are becoming more cautious about the endless financing of AI infrastructure.
So far, the largest technology corporations continue to assure the market that these projects will generate strong returns. Amazon CEO Andy Jassy has repeatedly expressed confidence in the future monetization of AI investments, while Meta CEO Mark Zuckerberg has described the expansion of computing infrastructure as one of the company’s most promising long-term priorities.
Nevertheless, the market is increasingly assessing not only potential earnings, but also the speed at which they can be realized. According to the companies themselves, Microsoft, Amazon, and Alphabet estimate that AI-related products and services could eventually generate as much as $1.45 trillion in revenue. However, even this amount remains below the total value of their financial commitments once hidden forms of financing are taken into account.
An additional concern is that the cost of such projects continues to grow rapidly after the start of construction. For example, Meta initially estimated the construction of a new data center in Louisiana at about $27 billion but later increased the project budget by almost half, to $50 billion.
Meta also relied on a financing structure that has become increasingly common in the AI infrastructure industry. Meta’s partner was the investment firm Blue Owl Capital, which acquired a stake in the joint venture, while Meta committed to leasing computing capacity once construction is completed. Formally, the company’s costs are lower than they would have been under direct ownership, although the economic substance of the project practically does not change.
Another risk is directly related to the profitability of AI infrastructure investments. If the pace of AI adoption turns out to be lower than expected or demand for computing power begins to weaken, the industry may face an oversupply of data centers, causing multibillion-dollar investments to take far longer to pay off than originally projected.
The recent experience of Intel illustrates these risks well. Until a few years ago, the company was considered one of the undisputed leaders of the semiconductor industry, but a combination of high capital expenditures, rising debt, and setbacks in the AI accelerator segment significantly weakened its financial situation. Today, Intel is undergoing a large-scale restructuring, with even some of its former competitors supporting the company. Despite these challenges, Intel’s stock has recently featured among market movers and premarket gainers.
As a result, the industry is gradually entering a new phase of development. Previously, investors viewed technology giants primarily as high-margin businesses requiring relatively little capital investment, but they increasingly resemble traditional infrastructure companies that must invest hundreds of billions of dollars annually to expand infrastructure.
So far, the market continues to believe that AI will provide sufficient revenue growth to justify these investments. However, as the scale of investments increases, so does the cost of making a mistake. That is why the upcoming quarterly reports from the largest technology companies will be evaluated by investors not only through the lens of business growth but also based on how effectively they manage their expanding financial commitments.
It’s a concern because the largest technology companies now carry trillions in visible and hidden commitments tied to AI infrastructure, which turns them into more capital-intensive businesses than investors are used to. As those obligations show up on balance sheets over time, markets will increasingly judge not just growth but how quickly AI revenues cover the rising debt load.
Hidden AI debt mostly takes the form of long-term, off-balance-sheet commitments to build or lease data centres and computing capacity that don’t appear as liabilities until facilities go live. The economic risk is real from day one, but the accounting treatment delays when it shows up in the headline numbers investors typically watch.
They’re changing perceptions because the spend levels now look more like traditional infrastructure than lightweight software, with hyperscalers planning hundreds of billions in capex each year. That shift means tech giants must continuously invest massive sums just to stay competitive, which raises the stakes if expected AI demand or pricing fails to materialise.
Strong AI revenue projections can help, but current estimates still sit below the total value of some companies’ combined on- and off-balance-sheet obligations. The gap forces investors to ask not only whether the revenue arrives, but whether it arrives fast enough and with high enough margins to justify the level of financial risk being taken.
If AI adoption or demand for computing power slows, the industry risks ending up with an oversupply of expensive data centres and hardware that take far longer to pay off than planned. In that scenario, high capex, rising debt and delayed returns could pressure earnings, valuations and future investment capacity across the sector.
