Big Tech’s AI Spending Test: When Investment Becomes Financial Risk
Microsoft, Alphabet, Amazon and Meta are committing extraordinary sums to artificial-intelligence infrastructure. Demand remains strong, but investors now need evidence that revenue, utilisation and cash generation can keep pace with the expanding cost base.
The artificial-intelligence investment boom has entered a more demanding phase.
For several years, investors largely rewarded technology companies for securing advanced chips, building data centres and establishing an early position in generative AI. Computing capacity was scarce, customer interest was accelerating and the largest platforms could not afford to fall behind.
That strategic argument remains persuasive. But the financial commitment has become too large to treat capital expenditure as a secondary line item.
Microsoft expects to invest approximately US$190 billion during calendar year 2026. Amazon has indicated that capital expenditure could reach about US$200 billion, predominantly to support Amazon Web Services and AI infrastructure. Alphabet previously forecast US$175 billion to US$185 billion before reportedly raising the range to US$195 billion to US$205 billion following its latest results. Meta entered 2026 expecting capital expenditure of US$115 billion to US$135 billion.
Together, the four companies are on course to invest more than US$700 billion in a single year.
The central question is no longer whether Big Tech believes in AI.
It is whether the economics of that investment will justify its scale and speed.
Demand is real
There is considerable evidence that AI infrastructure is already supporting revenue growth.
Microsoft said its AI business surpassed an annual revenue run rate of US$37 billion in its fiscal third quarter, representing year-on-year growth of 123%. Microsoft Cloud revenue reached US$54.5 billion for the quarter, while demand continued to exceed the company’s available computing capacity.
Alphabet has also reported strong demand. In its fourth quarter of 2025, Google Cloud revenue grew 48%, reaching an annualised run rate above US$70 billion. Cloud backlog increased to US$240 billion, driven partly by customer demand for AI products. The company said revenue from products built on its generative-AI models had risen by nearly 400% year on year during the quarter.
Amazon Web Services generated US$37.6 billion of sales during the first quarter of 2026, an increase of 28% from the previous year. Amazon has also said that customer commitments already cover a substantial portion of the AWS capacity it expects to build, although much of that capacity may not be monetised until 2027 or 2028.
Meta has a different route to monetisation. It does not primarily sell computing capacity to external cloud customers. Instead, it is using AI to improve advertising recommendations, user engagement and content discovery across Facebook, Instagram and its other applications. The company’s core advertising business therefore remains responsible for funding much of its infrastructure expansion.
These results suggest that the AI investment cycle is not based solely on speculative demand.
But strong demand does not automatically guarantee strong returns.
The accounting cost arrives gradually
Capital expenditure does not normally reduce reported profit immediately.
When a technology company purchases servers, networking equipment or constructs a data centre, the cash leaves the business upfront. The cost is then recognised gradually through depreciation over the asset’s estimated useful life.
This creates an important timing difference.
Revenue may begin increasing before the full accounting cost of the infrastructure becomes visible. During the early stages of an investment cycle, earnings can therefore appear resilient even as the company commits increasingly large amounts of cash.
The depreciation burden is now accelerating.
Alphabet said its depreciation expense rose 38% during 2025, from US$15.3 billion to US$21.1 billion. It expects depreciation growth to increase meaningfully in 2026, alongside higher energy and data-centre operating costs.
Microsoft reported that its company-wide gross margin percentage had declined because of continued investment in AI infrastructure and growing usage of AI products. Approximately two-thirds of its fiscal third-quarter capital expenditure went towards shorter-lived assets, mainly GPUs and CPUs.
This distinction matters.
A data-centre building may remain useful for decades. An advanced AI processor has a much shorter economic life and may need to be replaced as newer, faster and more energy-efficient chips become available.
Investors should therefore look beyond the headline capital-expenditure figure and ask three questions:
What type of asset is being purchased?
How quickly can it generate revenue?
How soon will it need to be replaced?
Free cash flow is the clearest pressure point
The largest technology companies can fund much of their AI investment from substantial operating cash flows.
That reduces the immediate risk of a conventional balance-sheet crisis. It does not remove the effect on shareholder economics.
Free cash flow is broadly the cash remaining after operating expenses and capital investment. It can be used for dividends, share repurchases, acquisitions or debt repayment.
When capital expenditure rises faster than operating cash flow, the amount available for those purposes falls—even when reported revenue and earnings continue growing.
Microsoft generated US$46.7 billion of operating cash flow during its fiscal third quarter. However, free cash flow was US$15.8 billion after higher capital spending. Microsoft still returned US$10.2 billion to shareholders through dividends and share repurchases, but its expanding infrastructure requirements are consuming a growing share of cash generation.
Amazon’s first-quarter cash capital expenditure increased from US$24.3 billion in 2025 to US$43.2 billion in 2026. The spending primarily reflected technology infrastructure for AWS and additional fulfilment capacity.
Meta generated US$115.8 billion of operating cash flow during 2025, but capital expenditure and finance-lease principal payments totalled US$72.2 billion. Its free cash flow for the year was US$43.6 billion. Meta nevertheless expects another major step-up in infrastructure investment during 2026.
None of these figures indicates imminent financial distress.
The more subtle risk is that future shareholder returns become increasingly dependent on AI revenue arriving on schedule.
Five tests for the AI investment cycle
Large capital expenditure is not necessarily a warning sign. It becomes more concerning when the evidence supporting future returns begins to weaken.
- Revenue growth relative to investment
Cloud and AI-related revenue should eventually grow sufficiently to justify the larger asset base.
If capital expenditure keeps accelerating while revenue growth slows, returns on invested capital may deteriorate—even when absolute revenue continues rising.
- Capacity utilisation
Management teams continue to report that demand exceeds available supply. That supports further investment.
The more difficult test will come when the industry moves from scarcity to abundance. Data centres with insufficient customers or underused GPUs would turn a strategic advantage into an expensive fixed cost.
- Gross margins
AI products can generate substantial revenue while still placing pressure on margins because inference, energy and infrastructure costs are high.
A temporary decline in gross margin may be reasonable during a large investment cycle. Persistent deterioration could indicate that competition is limiting pricing power or that customers are capturing more of the economic value than the infrastructure providers.
- Depreciation and replacement cycles
The useful lives assigned to servers and other equipment affect reported depreciation.
Longer accounting lives reduce annual depreciation expense, but they do not change the economic reality if hardware becomes obsolete sooner than expected.
Investors should compare reported asset lives with the pace of technological improvement.
- Customer concentration
Not all contracted demand carries the same risk.
A diversified base of paying cloud customers is more resilient than infrastructure built around a small number of large AI laboratories or strategic partners.
Microsoft’s remaining performance obligations reached US$627 billion in its fiscal third quarter, but the company noted that this figure included significant OpenAI commitments.
Large contracts can offer visibility. They can also create customer-concentration and counterparty risks.
The risk may emerge outside Big Tech first
Microsoft, Alphabet, Amazon and Meta entered this investment cycle with profitable core businesses, substantial cash flow and access to global capital markets.
That makes the current build-out different from speculative infrastructure booms financed almost entirely through fragile debt.
However, the broader AI ecosystem may carry more financial leverage than the technology companies’ headline accounts suggest.
Data-centre developers, utilities, chip suppliers, private-credit funds and specialised financing vehicles are all contributing to the expansion. Finance leases also allow companies to secure capacity while spreading some cash payments over time.
Microsoft recorded US$4.7 billion of finance leases during its fiscal third quarter, mainly for large data-centre sites. It expects interest payments related to those leases to contribute to interest expense.
Amazon has explained that it may need to commit capital for land, power, buildings and equipment well before customer billing begins. This introduces a timing risk: the investment is made today, but the revenue may arrive several years later.
Financial stress could therefore appear first among highly leveraged infrastructure providers, utilities or data-centre developers rather than at the largest technology platforms themselves.
What to Watch
The next stage of the AI investment cycle should be judged through several indicators.
Cloud growth and backlog: Azure, Google Cloud and AWS need to keep converting customer commitments into recognised revenue.
Capital-expenditure guidance: Further increases may reflect genuine demand, but each revision raises the level of future revenue required to earn an acceptable return.
Depreciation growth: Rising depreciation will reveal more of the economic cost of infrastructure installed during previous years.
Free-cash-flow conversion: Operating cash flow must expand sufficiently to offset higher capital requirements.
AI product monetisation: Investors need clearer evidence of how copilots, agents, subscriptions, advertising improvements and custom chips are contributing to profits—not merely usage.
Capacity constraints: Shortages currently support utilisation and pricing. A move towards excess capacity could alter the economics quickly.
Financing structures: Greater reliance on leases, private credit or off-balance-sheet arrangements would warrant closer scrutiny.
Finance World’s Read
Big Tech’s AI investment boom is not yet a balance-sheet crisis.
The leading platforms have real customer demand, powerful existing businesses and the financial capacity to fund an infrastructure cycle that few competitors can match.
But strategic importance should not be confused with guaranteed financial returns.
The spending becomes a material risk when investment grows faster than monetisation, utilisation weakens, or depreciation and operating costs begin eroding margins without a corresponding increase in durable revenue.
For investors, the most important question is no longer which company is spending the most.
It is which company can convert computing capacity into recurring, high-margin cash flow before the assets need to be replaced.
Sources
Microsoft, fiscal 2026 third-quarter earnings conference call.
Alphabet, fourth-quarter 2025 earnings call.
Amazon, first-quarter 2026 earnings release and Form 10-Q.
Amazon, 2025 shareholder letter and 2026 capital-expenditure commentary.
Meta, fourth-quarter and full-year 2025 results.
Reuters, Alphabet’s July 2026 earnings and revised capital-expenditure guidance.