The AI spending boom has a shadow. Nine of the largest technology companies are sitting on approximately $3 trillion in off-balance-sheet commitments tied to artificial intelligence infrastructure, according to a Wall Street Journal analysis. That figure is roughly five times the $600 billion in combined capital expenditures these firms reported over their most recent 12-month periods.
Where the money is hiding
The WSJ breakdown splits the $3 trillion into two main buckets. The first is unstarted leases, valued at roughly $904 billion to $1.2 trillion depending on the data source. These are commitments to data-center space that companies have locked in but haven’t yet begun using, meaning they don’t show up as liabilities on the balance sheet under current accounting rules.
The second bucket is purchase commitments, estimated between $1.52 trillion and $1.9 trillion. These cover long-term agreements for chips, data-center construction, and energy procurement.
Individual company figures paint an even starker picture. Alphabet leads the pack with $811 billion in purchase and contractual commitments as of June 30, 2026. Meta, meanwhile, carries $347 billion in leases that haven’t yet commenced according to its latest filings.
These numbers come directly from the companies’ own SEC filings, just not from the parts most investors typically read. The obligations live in footnotes, the financial equivalent of fine print on a rental car agreement.
The accounting reality
Under current accounting standards, leases that haven’t commenced don’t appear as liabilities on a company’s balance sheet. Purchase obligations similarly get different treatment than outright capital expenditures. The result is that traditional metrics like debt-to-equity ratios and reported leverage don’t capture the full scope of what these companies have committed to spend.
The WSJ analysis wasn’t the first to flag this issue. A Nikkei report published in July 2026 estimated $1.65 trillion in off-balance-sheet AI debt across just five companies: Alphabet, Amazon, Meta, Microsoft, and Oracle. The WSJ’s broader nine-company scope nearly doubled that figure.
Investor Michael Burry, who famously bet against the housing market before the 2008 financial crisis, noted the WSJ findings and claimed to have identified similar risks as early as 2025.
What this means for markets and investors
The comparison to reported capex matters because it reveals how much of the AI infrastructure buildout is essentially financed through future obligations rather than current spending. At $600 billion in trailing 12-month capex, these companies are already spending at historically aggressive rates. The $3 trillion in additional commitments suggests the real number is far larger.
For equity investors, valuations that look reasonable based on reported financials may need recalibrating once off-balance-sheet commitments enter the picture. Analysts who model free cash flow without accounting for these future outflows could be significantly overestimating the cash available for dividends, buybacks, and other shareholder-friendly activities.