AI infrastructure spending is pushing the largest technology companies toward debt, equity and customer-backed financing, raising new questions about cash flow, capital discipline and the returns on record investment.
For years, the largest technology companies financed expansion from the cash generated by advertising, cloud computing, software and e-commerce. The artificial intelligence buildout is testing that model. The sums required to build data centers, acquire advanced chips and expand computing capacity are growing so quickly that even highly profitable companies are turning increasingly to external capital to fund their ambitions.
Alphabet’s second-quarter results illustrate the pressure. The company spent $44.9 billion on capital expenditure, leaving it with negative free cash flow of $5.9 billion. Free cash flow measures the cash left after operating expenses and capital investment. Alphabet generated $39.1 billion in operating cash flow during the quarter, but that was not enough to cover its infrastructure spending.
The result does not, by itself, indicate financial distress. It does show how quickly capital-intensive AI investment can absorb cash generated by an established business. The larger question is whether the revenue and operating cash flow generated by new infrastructure will eventually justify the scale of investment required to build it.
Across Alphabet, Amazon, Meta, Microsoft and Oracle, the five companies commonly grouped as hyperscalers, capital spending is expected to exceed $690 billion in fiscal 2026, according to FactSet. Over the past 18 months, the group has moved from relying predominantly on internally generated cash toward a wider mix of debt, equity and customer-backed arrangements.
The investment bill keeps climbing
The scale of the commitment continues to rise. Three of the four largest spenders raised their 2026 capital=spending forecasts during the latest earnings round, highlighting the difficulty of predicting how much infrastructure the AI market will require.
Alphabet Inc. raised its 2026 capital expenditure forecast to $195 billion to $205 billion from $180 billion to $190 billion. CFO Anat Ashkenazi said the company is delivering capacity faster to meet demand.
Amazon’s estimate rose to about $220 billion from $200 billion. CFO Brian Olsavsky pointed to higher memory costs, which means that rising capital expenditure can reflect both additional infrastructure and more expensive equipment. The distinction matters because a larger spending budget does not necessarily translate into a proportionate increase in computing capacity.
Meta narrowed its range to $130 billion to $145 billion; its implied second-half spending of $79 billion to $94 billion would be nearly double the $50.9 billion it spent in the first half.
Microsoft’s number moved the other way, to about $175 billion from $190 billion, but that reflects a lease-accounting change. CFO Amy Hood said its spending plans remain unchanged.
Taken together, the forecasts point to an industry still prioritizing capacity expansion. They also show why headline spending figures need context: changes in accounting treatment, equipment prices and project timing can affect reported capital expenditure without necessarily indicating a change in strategic intent.
Debt and equity fund a growing share of the buildout
The growing investment bill is changing how the companies finance expansion. According to FactSet, incremental annual debt rose from 9% of capital spending in fiscal 2024 to 32% over the 12 months through mid-2026. FactSet says most of the five companies still have strong balance sheets and low debt levels.
That financial strength gives them room to borrow. But the increasing use of external financing changes the economics of the buildout: capital investment must now be assessed alongside debt obligations, potential shareholder dilution and the cash available for other corporate priorities.
Alphabet illustrates how several funding sources can operate in parallel. Its debt reportedly rose from about $16 billion to roughly $100 billion in a year. In June, the company announced an equity capital raise initially valued at $80 billion, including a $10 billion private placement with Berkshire Hathaway. The announced package included public offerings, mandatory convertible preferred stock and an at-the-market share offering program.
The financing structure matters. The $80 billion headline was not simply a pool of cash earmarked entirely for AI construction. Alphabet said proceeds from its public offerings and private placement would support general corporate purposes, including AI infrastructure and computing capacity. It also said the at-the-market program was intended primarily to help meet tax obligations associated with employee equity awards. The company subsequently announced an increase in the proposed total equity raise to $84.75 billion.
Alphabet’s finance chief has warned that free cash flow “will remain under pressure.” The company is therefore expanding infrastructure while adding debt and equity financing to its funding mix.
Meta presents a different version of the same challenge. It generated $31.9 billion in operating cash flow but reported free cash flow of just $784 million. It also borrowed $24.9 billion, stopped share repurchases and brought in a BlackRock-backed venture for a 1-gigawatt data center in El Paso, Texas.
Microsoft, by comparison, generated $19.6 billion in free cash flow and expects to remain free-cash-flow positive through fiscal 2027. That contrast is important: the hyperscalers share a broad investment cycle, but they do not have identical cash-generation profiles or financing requirements. Their ability to fund AI infrastructure internally will depend on their existing businesses, investment schedules and other capital commitments.
Returns arrive over a longer horizon than costs
AI infrastructure creates a mismatch between when companies spend and when they expect to recover the investment. Data centers, power infrastructure and computing equipment require substantial upfront capital. Revenue accumulates over time as customers use the capacity, while depreciation, maintenance, financing and operating costs continue.
Amazon’s cash flow illustrates the effect. Its trailing 12-month free cash flow turned negative, reaching an outflow of $7.6 billion in the second quarter. Operating cash flow, meanwhile, rose 33% to $161.4 billion. The company attributed the free-cash-flow decline to higher capital expenditure rather than weaker underlying business performance.
Chief Executive Andy Jassy has argued that data centers can generate revenue for more than 30 years, while servers and networking equipment can break even in less than three years against a useful life of five to six years.
That is the investment case from management’s perspective: the initial outlay is large, but the infrastructure can support revenue over an extended period. The risk is that the expected economics depend on sustained demand, effective utilization and the ability to generate enough revenue from each unit of capacity. A long-lived facility does not guarantee that every generation of computing equipment will remain equally competitive or that customer demand will match the capacity being built.
The financial question is not simply whether AI demand is growing. It is whether revenue and cash generation can grow fast enough to support the infrastructure, including when equipment must be upgraded and new capacity added.
Customer commitment help fund the buildout
Borrowing and equity issuance are only part of the financing picture. Customer commitments can also help companies manage the cost and risk of expansion, particularly when large customers reserve capacity or supply some of their own hardware.
Oracle has used customer prepayments and “bring your own hardware” arrangements, under which clients supply their own chips, to limit its upfront cash requirements. The company reported $75 billion in hardware contracts in its latest earnings update, according to the figures supplied. Its remaining performance obligations, which measure contracted revenue not yet recognized, rose 363%.
Oracle plans to raise about $40 billion through debt and equity in fiscal 2027.
Microsoft’s backlog grew 84%, according to the supplied figures, but the increase was 25% excluding its OpenAI contract. These commitments offer evidence of demand, but they need to be interpreted carefully. Backlog is not the same as revenue already recognized, and customer agreements do not eliminate the costs of building and operating infrastructure. The timing of delivery, contractual terms and the conversion of commitments into cash remain material to the financial outcome.
Customer-backed arrangements can reduce the amount of capital a provider must commit before a project advances. They can also create dependence on a smaller number of large customers. That concentration becomes more consequential when infrastructure investment is designed around specific contracts or customer requirements.
Credit risk and capital discipline come into focus
Oracle’s financing needs have attracted particular scrutiny. S&P Global Ratings cut the company’s credit rating to BBB-minus, the lowest investment-grade tier, from BBB on July 9. The agency cited rising capital spending, negative free cash flow and customer concentration.
The downgrade highlights a distinction between the ability to raise capital and the cost of doing so. A company may retain access to debt markets, but a weaker credit profile can make financing more expensive and place greater emphasis on future cash generation.
The opportunity cost of AI spending also matters. FactSet has warned that weaker free cash flow could constrain share repurchases and research and development outside AI. If the expected returns fail to materialize, management teams may face pressure to reconsider the pace of expansion or make trade-offs between AI infrastructure and other investments.
For now, the companies have different levels of flexibility. Strong existing businesses and balance sheets can absorb periods of elevated capital expenditure. But the more spending relies on borrowing or new equity, the more important it becomes to demonstrate that the investment can produce sustainable economic returns.
The spending cycle may extend into 2027
Management commentary suggests that the buildout is far from complete. Amazon’s Jassy said capacity constraints would likely persist through 2027. Oracle indicated capital spending of roughly $70 billion in fiscal 2027, while Alphabet signaled further increases. Meta and Microsoft did not provide specific capital-spending forecasts, and Meta Chief Financial Officer Susan Li described capital planning as “highly dynamic.”
Nvidia, a major supplier of chips used in AI infrastructure, offered an optimistic view. Chief Financial Officer Colette Kress cited third-party forecasts that hyperscaler spending could exceed $1 trillion in 2027.
Those forecasts underline the scale of the industry’s ambitions, but they remain projections rather than evidence of realized returns. Continued spending will depend on demand for computing capacity, the economics of AI services, access to power and equipment, and the ability of companies to bring projects online as planned.
The funding shift does not mean the AI investment cycle is unsustainable. Nor does rising demand guarantee that every infrastructure project will deliver the returns its backers expect. The five companies enter this phase with different cash flows, capital structures and customer commitments.
The decisive measure will be what the investment produces. If AI-related revenue and operating cash flow grow sufficiently to cover ongoing capital needs and financing costs, today’s spending may support durable businesses. If returns lag behind the buildout, companies will have to make harder choices about borrowing, shareholder payouts and the next round of investment. The test is moving beyond how much infrastructure Big Tech can build to how effectively it can turn that capacity into cash.
