Tracking Off-Balance Sheet Lease Commitments for Data Centers

Everyone knows that the tech giants and AI hyperscalers are betting big on data centers. Calcbench has been taking a deep dive into Q2 disclosures of those companies, and today we offer a better sense of how big those bets are.

Bets, by the way, that aren’t included on the tech giants’ balance sheets.

These bets travel under the rather boring names “uncommenced leases” or “unrecognized lease commitments.” As the name implies, these are leases (typically for AI data centers) that the company has signed and do exist, but the leases haven’t yet started and don’t appear on the company’s balance sheet. 

Altogether, among the six companies leading the charge on data center development, these uncommenced lease expenses now exceed $1 trillion.

Figure 1, below, shows how the expenses have soared in recent years — from $321.5 billion in 2024, to $700.2 billion in 2025, to $1.13 trillion as of Q2 2026. 

As you can see, different companies are racking up these uncommenced lease costs at different rates. In relative terms, the one with the steepest increases is Google Alphabet ($GOOG), which had a jump of 1,150 percent; but that’s mostly because Alphabet started from an extremely low amount in 2024 ($7.2 billion), which reached $91 billion today. But that $91 billion is still lower in absolute dollars than the uncommenced lease commitments carried by Facebook Meta ($META) or Microsoft ($MSFT), which currently stand at $279 billion and $329 billion, respectively.

Another way to think about the numbers is to look at which company was incurring the largest share of unrecognized commitments in any given year. That is, if all six companies had $100 billion in unrecognized commitments in 2025, each company accounted for how much of that total? That’s represented in Figure 2, below. 

As you can see, Facebook has been accounting for an increasingly large percentage of the whole amount every year, even as that whole amount grew larger in absolute dollar terms year over year. 

Finally, we did a quick analysis to compare these off-balance sheet commitments to each company’s total liabilities. That gives a sense of how much the company’s balance sheet could go through the blender if those off-balance sheet commitments suddenly did have to be brought back onto the balance sheet. See Figure 3, below.

If any readers are suddenly wondering, “Wait, isn’t this what happened with Lehman Bros. in 2008 just before the financial crisis?” — well, it could be. 

For example, if Facebook had to bring all those not-yet recognized lease commitments onto the balance sheet all at once, without any corresponding increase in assets because nobody was using AI like forecasters expected, that would balloon total liabilities by 148 percent and be a disaster for stockholders. But Facebook claims it does have revenue commitments to back up all these lease commitments when the time comes. If those revenue commitments turn into actual revenue, then everything will be fine.

Where to Find All These Disclosures

That’s easy enough. For starters, you can always use the Calcbench Disclosures & Footnotes Query page to pull up specific footnote disclosures and read exactly what the company is saying. The good stuff is always in the fine print!

That said, not all companies disclose their unrecognized leasing commitments in the same location. For example, Amazon ($AMZN) discloses lease information in its Commitments and Contingencies footnote, because leases are commitments to future expenses. In contrast, Microsoft reports its future costs in a dedicated Leases footnotes, because the commitments are leases.

So you need to look. You can do that by studying the exact footnote disclosures that each company makes from the list on the left-hand side of your screen and then choosing whichever footnote makes the most sense. You might need to search both the Commitments and Leases footnotes if a company reports them both, but the information will be in there somewhere.

You can also use our Multi-Company page to search disclosures across a group of companies. Start by searching for the XBRL tag:

UnrecordedUnconditionalPurchaseObligationBalanceSheetAmount

That should pull up the relevant amounts for whatever period you’re searching. You can then do a time-series pull to see how that amount has changed over time and export the whole thing in Excel.