Have you noticed how the biggest technology companies keep posting eye-watering capital spending numbers while their free cash flow quietly turns negative? I have been watching this pattern for months, and the part that still surprises me is how little of the true commitment shows up on the balance sheet. The visible capital expenditure figures already run into the hundreds of billions each year. What sits off the books is larger still, and it keeps expanding at a pace that feels almost unreal.
The Hidden Scale Of AI Infrastructure Promises
Across the major cloud and platform operators, the total of signed but not yet started leases plus long-term purchase obligations has climbed past the three-trillion-dollar mark. That figure alone is roughly five times the amount of traditional capital expenditure these same companies reported over the past twelve months. In a single recent quarter the off-balance-sheet total jumped by more than a trillion dollars. When numbers move that fast, the usual quarterly updates start to feel incomplete.
Most of the growth comes from two sources. First are the multi-year leases for data-center space that have been signed but have not yet begun. Accounting rules keep those future rent payments off the balance sheet until the buildings open and the rent checks start flowing. Second are the purchase commitments for chips, servers, networking gear, and power equipment. Those contracts also stay off the books until the hardware is delivered. Both practices are completely standard. The issue is the sheer size and the speed at which the totals are rising.
How Uncommenced Leases Create Invisible Obligations
Take one large social-media and advertising company as an example. It has already committed to a data-center campus measured in thousands of football fields. The initial lease term is short, yet the company has guaranteed bondholders that it will stay for up to twenty years. Because management does not currently expect to walk away, no liability appears on the balance sheet. The disclosed aggregate commitment for leases that have not yet started nevertheless sits in the hundreds of billions. Similar arrangements appear across the rest of the group of large spenders.
I keep coming back to the same practical question: what happens if demand softens before those leases kick in? The contracts are largely non-cancellable. The companies will still owe the rent. In the optimistic view, artificial-intelligence usage keeps expanding fast enough to fill every rack. In the more cautious view, the industry ends up with expensive capacity that is only partially used. Either way, the cash obligation remains.
Purchase Commitments And The Chip Supply Chain
Hardware contracts form the other large piece. Leading chip designers and cloud providers have locked in production capacity years ahead of actual delivery. Those commitments total close to two trillion dollars across the group of companies that disclose the figures. Under ordinary accounting they stay off the balance sheet until the products arrive. That treatment is correct, yet it also means the true capital intensity of the AI build-out is systematically understated in the headlines that focus only on reported capital expenditure.
The concentration of these deals creates another layer of risk. A handful of suppliers and a small number of private-credit arrangements sit at the center of the funding structure. If one major counterparty faces stress, the effects can ripple through several balance sheets at once. Think of it as a modern version of rehypothecated leverage, only this time the underlying asset is future compute capacity rather than securities.
Why Free Cash Flow Turned Negative
Two of the largest platform companies recently reported free cash flow that was negative for the period. In plain language, the cash they spent on infrastructure exceeded the cash generated by their ongoing operations. That shift would have been almost unthinkable a few years ago when these same firms were famous for fortress balance sheets and massive cash piles. The change is not temporary. Management teams have signaled that elevated spending will continue for the foreseeable future.
Once you add the off-balance-sheet commitments to the picture, the funding requirement looks even larger. The companies have been raising debt and arranging private-credit facilities while market sentiment remains favorable. That window will not stay open forever. The practical question is whether the cash arrives in time to meet the obligations already signed.
Depreciation Is Still Waiting In The Wings
Another timing mismatch is already visible in the construction-in-progress balances. Those accounts have grown dramatically, in some cases more than doubling year over year. The depreciation expense associated with all the new equipment has not yet hit the income statement in full. When it does, the impact will be material. For several of the largest spenders, cumulative depreciation over the next three years could exceed half a trillion dollars. As a share of revenue the figure is expected to rise sharply.
To keep operating margins stable, either revenue has to grow in lockstep or other cost lines have to fall. So far the revenue growth, while strong, has not fully kept pace with the capital commitments already locked in. That gap is one reason the off-balance-sheet totals matter. They represent future cash outflows and future depreciation that are already contracted but not yet fully visible in the reported numbers.
The hazard is not fraud. It is timing and visibility. Capex is being committed ahead of the revenue and free cash flow meant to support it.
Disclosure Differences Make Comparison Difficult
Even careful readers of the filings face an extra obstacle. Companies do not classify or quantify the same obligations in the same way. One firm may break out cloud-service commitments in detail while another folds them into a larger purchase-obligation number without further breakdown. Whether a particular capacity deal counts as a lease can turn on judgment calls about control of the hardware and the workloads. Those judgment calls produce inconsistent disclosures, which in turn make cross-company comparisons of true leverage less precise than they appear.
I have spent enough time with the footnotes to know that the aggregate three-trillion-dollar figure is itself an estimate built on differing materiality thresholds. The real exposure could be somewhat higher or somewhat lower. The direction of travel, however, is unambiguous. Every recent quarterly update has pushed the total higher.
Optimistic And Cautious Scenarios
There is a coherent optimistic case. Demand for artificial-intelligence tools continues to climb. Hardware shortages persist. Token usage grows fast enough that the new capacity is filled at attractive prices. In that world the companies generate the cash needed to meet every obligation, and the off-balance-sheet totals simply become the foundation of a larger, more profitable franchise.
The cautious case focuses on price pressure and shifting competitive dynamics. Open-weight models developed outside the traditional Western supply chain have already driven token prices sharply lower in a short period. If that trend continues, the economics of frontier model training and inference could change. Capacity that looked essential at today’s prices might look excess at tomorrow’s prices. The lease and purchase contracts would still have to be honored.
Neither scenario is certain. What is certain is that the obligations already exist. They cannot be canceled for the most part. The companies that once prided themselves on pristine balance sheets now carry a growing stack of future payments that will arrive whether or not the revenue materializes on the expected schedule.
Private Credit And Circular Funding Arrangements
One recent development illustrates how the funding side is evolving. A leading chip supplier formalized a large open-ended arrangement with a group of private-credit firms. The structure is designed to support the very same data-center and hardware build-outs that sit behind the off-balance-sheet totals. Similar arrangements have appeared before, but the scale is new. The private-credit layer now sits between the hyperscalers and the ultimate providers of capital.
That intermediation has benefits. It spreads the financing burden and keeps some of the leverage off the technology companies’ own balance sheets. It also creates new points of potential stress. If credit markets tighten or if one of the private-credit vehicles faces losses, the funding pipeline for the next wave of capacity could narrow quickly. The industry is already relying on continued access to that pipeline.
What The Numbers Mean For Investors
For equity investors the immediate implication is that traditional valuation metrics based on reported capital expenditure understate the true capital intensity of the business. Free-cash-flow yields look higher than the underlying commitment structure would suggest. Debt ratios calculated from the face of the balance sheet miss the leases and purchase obligations that will convert into cash outflows and, eventually, into reported liabilities.
Credit investors face a related issue. The companies still enjoy strong credit ratings and deep access to the bond market. The question is how those ratings will evolve once the depreciation wave arrives and once the off-balance-sheet items begin to migrate onto the balance sheet. Rising leverage ratios and higher interest expense would be the natural consequence if revenue growth slows.
I do not claim that a crisis is imminent. The current risk is more subtle. It is a set of timing mismatches, disclosure inconsistencies, and concentrated counterparty exposures that will matter most if sentiment shifts. The companies themselves appear to understand the window of opportunity. They are raising capital while conditions remain favorable. The open question is whether the capital raised will prove sufficient once the funding environment becomes less accommodative.
Looking Ahead At The Funding Question
The next several quarters will show whether the off-balance-sheet totals continue their rapid climb or begin to stabilize. They will also reveal how quickly the depreciation expense flows through the income statements. Most important, they will test the assumption that demand for high-end compute remains robust enough to justify every commitment already signed.
In my own reading of the filings and the broader market commentary, the most interesting tension is not between bulls and bears in the usual sense. It is between the speed of capital commitment and the speed of monetization. Capex revisions have consistently run ahead of sales and free-cash-flow revisions. That pattern can persist for a while. It cannot persist indefinitely without either stronger revenue growth or some form of capacity adjustment.
The three-trillion-dollar figure is already large enough to command attention. The fact that it grew by more than a trillion dollars in a single quarter makes the trajectory even harder to ignore. Whether one views the build-out as the foundation of a new computing era or as an over-commitment that will eventually require painful rationalization, the scale of the off-balance-sheet liabilities is now an essential part of the story.
Investors who continue to focus only on the capital-expenditure line in the cash-flow statement are looking at an incomplete picture. The fuller picture includes the leases that have not yet started, the purchase orders that have not yet been delivered, and the depreciation that has not yet been recognized. Those items are already contracted. They will arrive on their own schedule, regardless of how the narrative around artificial intelligence evolves in the months ahead.
That, more than any single quarterly number, is the point worth watching. The industry has chosen to place an enormous bet on future demand. The terms of that bet are increasingly visible in the footnotes rather than on the face of the balance sheet. Understanding those footnotes has become a necessary part of evaluating the durability of the current investment cycle.
