Nine of the largest U.S. technology companies have disclosed roughly $3 trillion in future commitments, largely connected to artificial intelligence infrastructure, underscoring how the cost of the AI buildout is extending far beyond the capital spending already visible on corporate balance sheets.
Alphabet, Meta, Microsoft, Amazon, Oracle, Nvidia, Broadcom, SpaceX and AMD collectively reported commitments that include an estimated $1.2 trillion in uncommenced leases and about $1.9 trillion in purchase obligations, according to their latest disclosures. The total is about five times the group’s combined $600 billion in capital expenditures over the past year and roughly three times their combined lease liabilities and long-term borrowings.
These commitments are disclosed in financial-statement notes, but much of the spending does not yet appear as a balance-sheet liability under existing accounting rules. That treatment allows companies to reserve data-center capacity, secure long-term chip supplies and arrange specialized financing years before the related equipment is delivered or lease payments begin.
The aggregate figure climbed from around $1.8 trillion to $3 trillion in roughly two months, based on the reported disclosures. The pace of that increase reflects the scale of contracts being signed as major technology companies compete for computing power needed to train and operate large AI models.
Leases lock in future data-center costs
Uncommenced leases account for an estimated $1.2 trillion of the total, about four times the level disclosed a year earlier. Such leases generally remain outside a company’s recognized lease liabilities until the underlying facility is available for use and rent payments begin.
That accounting distinction is especially relevant for large AI campuses, where construction and energy arrangements can take years to complete. Companies can commit to occupy facilities well before they begin using the computing capacity housed inside them.
Meta disclosed about $347 billion in uncommenced lease commitments. Its filings included the Hyperion data-center project in Louisiana, described as spanning an area comparable to roughly 1,700 football fields.
The Hyperion lease is expected to start in 2029 with an initial four-year term, according to Meta’s disclosures, followed by renewal options that could extend the relationship to 20 years. Meta also described an obligation to cover potential bondholder shortfalls if the lease were ended early, creating an additional contingent exposure around the project’s financing.
For companies building AI infrastructure, these arrangements can provide access to specialized campuses without directly funding every stage of construction. The trade-off is that long-duration leases may become fixed costs well before the revenue from AI products, cloud services or enterprise contracts is clear.
Supply contracts add nearly $2 trillion
Purchase obligations, estimated at about $1.9 trillion across the nine companies, form the largest part of the commitments. These agreements can cover long-term supplies of data-center hardware, including AI chips, memory and related systems.
The contracts help technology groups secure capacity in a market where leading processors and high-bandwidth memory have become strategic bottlenecks. For chip designers and suppliers, long-term commitments improve demand visibility and can support investment in manufacturing and packaging capacity.
For customers, the agreements create a different form of risk: many are largely non-cancellable once signed. A company that overestimates demand for AI computing services could still face large hardware and infrastructure payments even if usage, pricing or profit margins disappoint.
Alphabet and Amazon have recently reported negative free cash flow, with capital expenditures exceeding operating cash inflows, according to the financial figures cited in the disclosures. Future commitments add another layer to that spending picture because they represent planned obligations that may convert into cash outflows in later years.
The issue is less about whether these companies can currently fund their commitments than about how quickly obligations are accumulating relative to the still-developing economics of generative AI. Cloud providers and model developers are spending ahead of a settled answer on how much customers will pay for inference, training and AI-enabled software.
Residual-value guarantees shift part of the risk
A separate financing structure, known as a residual-value guarantee, or RVG, has also emerged around AI hardware projects. The structure was estimated at roughly $70 billion in recent market reporting.
Under an RVG arrangement, a special-purpose vehicle may borrow money to acquire chips, then rely on customer usage contracts, resale proceeds or re-leasing income to repay lenders. A guarantee can cover any remaining shortfall if the equipment’s value or customer revenue falls below expectations.
Nvidia and Broadcom have been associated with RVG-style support. In such structures, a company may not record a liability if it assesses payment under the guarantee as unlikely. Meta’s filings said RVG guarantors had not recorded liabilities because payment was not considered probable.
Broadcom was linked to the Big Sky project, where it provided support for a $35 billion debt transaction involving Apollo Global Management and Blackstone. The financing was used to fund custom AI chips leased to Anthropic, placing a major hardware supplier alongside private-credit firms and an AI developer in the same capital structure.
Bank of America strategists estimated that Broadcom’s AI XPV platform could accumulate $370 billion in senior debt by mid-2029. Nvidia Chief Executive Officer Jensen Huang has also discussed potential residual-value support of as much as 25% in selected cases, while describing a $500 billion financing cooperation with six U.S. investment firms, including BlackRock and Goldman Sachs.
Rating agencies focus on contingent obligations
Credit-rating agencies have started to flag these guarantees as exposures that can matter even when reported debt remains low.
Moody’s said a rapid rise in contingent obligations could reduce Broadcom’s financial flexibility, particularly if many transactions are completed in a short period. S&P Global Ratings has categorized Broadcom’s residual-value support as a contingent debt-type obligation and said it would include the support in its adjusted debt calculations.
That treatment illustrates the gap between standard accounting presentation and credit analysis. A lease or guarantee may sit outside reported debt today, yet lenders and rating agencies can still treat it as a future claim on a company’s cash flow.
The AI infrastructure race is therefore tying the sector’s largest companies to substantial future payments through leases, supply contracts and financing guarantees. The strategy can secure scarce computing capacity and accelerate new data-center construction, but it also leaves company budgets more exposed if AI revenue grows more slowly than the hardware commitments now being signed.
Track how mega AI investments shape crypto liquidity and trading trends in 2026 with our latest market outlook.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
