For a decade the largest technology companies were defined by cash generation so abundant that the main question was what to do with it. In 2026 the question changed. Capital spending has grown fast enough that free cash flow is compressing at companies that once seemed structurally incapable of running short of money.
This article is for informational and educational purposes only. It is not financial advice, investment advice, or a recommendation to buy, sell, or hold any security, cryptocurrency, or financial product. Always verify data with official sources before making financial decisions.
The ratio that changed
Short answer: Free cash flow is operating cash flow minus capital expenditure. The hyperscalers have historically converted a large share of operating cash flow into free cash flow because their capital intensity was low relative to their profitability. AI infrastructure spending has raised capital expenditure sharply, and free cash flow has compressed as a result — at some companies substantially, even while operating cash flow continued to grow.
Operating cash flow vs capex
The useful framing is capital intensity: capital expenditure as a share of revenue or of operating cash flow. A software business converting most of its operating cash flow to free cash flow is a fundamentally different asset from one reinvesting the majority of it in physical infrastructure.
That second profile resembles a utility or a telecom operator more than it resembles the software businesses these companies were valued as. The valuation implication is not automatic — high reinvestment at high returns creates value — but the multiple appropriate to a business with heavy ongoing capital needs differs from one that requires almost none, and the market has been working through that repricing.
Finance leases and the capex that is not in the capex line
This is the disclosure most commonly missed in capital spending analysis.
Not all infrastructure is purchased outright. Data centre capacity is frequently acquired through leases, and finance leases put an asset and a corresponding liability on the balance sheet while the cash outflow appears in financing activities rather than in capital expenditure.
The consequence is that headline capital expenditure can understate total infrastructure commitment. Companies disclose finance lease additions in the cash flow statement and in the leases note, and some now provide a combined figure precisely because the headline number was becoming misleading. Comparing capital expenditure across companies without adjusting for lease treatment compares different accounting choices rather than different spending levels.
Debt issuance by companies that did not need it
Several of these companies carry large cash balances and have historically had little use for debt markets. Increased issuance is therefore notable — not because the debt is burdensome relative to their balance sheets, but because it indicates that internally generated cash is no longer sufficient to fund the spending programme at the desired pace.
There is a second, less discussed motivation: much of the cash sits offshore or is committed, and issuing debt against a strong credit rating at manageable rates can be cheaper than repatriating or liquidating investments. Both explanations are consistent with the observed issuance, and they have different implications for how constrained these companies actually are.
What happens to buybacks when free cash flow compresses
These companies have been among the largest repurchasers of their own stock, which made them a significant source of demand for their own shares and, given index concentration, for the market.
Buybacks are discretionary and are the first thing reduced when capital is needed elsewhere. Announced authorisations are not commitments — they are permissions, and actual repurchase activity disclosed quarterly frequently runs below the authorised pace. Watching actual repurchases rather than announcements is the only way to see this, and it appears in the cash flow statement.
The second-order effect matters given index concentration: reduced buying by the largest index members removes a source of price-insensitive demand from the most heavily weighted part of the market.
The bull case: capacity, not stranded assets
The counterargument deserves proper weight rather than a dismissive sentence.
This capital spending buys revenue-generating capacity. Cloud infrastructure has historically produced strong returns on invested capital, and the companies building it are doing so against contracted demand, not speculatively. If AI workloads generate the revenue expected, high capital intensity now produces a larger, more profitable business later, and the current free cash flow compression is an investment phase rather than a deterioration.
The bear case is not that the spending is irrational but that it is defensive — that competitive dynamics compel each participant to spend regardless of returns, because falling behind is worse than overspending. Both readings are consistent with the observed numbers, and the evidence that would distinguish them is several years of revenue data that does not exist yet.
What this article does not conclude
Nothing here forecasts returns on this capital spending or any company’s free cash flow. Capital expenditure figures and lease treatment vary by company and by disclosure practice.
Cash flow statements, lease notes and repurchase disclosures are in the quarterly and annual filings. Aggregated capital spending figures circulating in commentary frequently mix companies with different fiscal years and different lease accounting, and are not directly comparable.