Microsoft, Alphabet, Meta and Amazon have given investors the first substantial evidence that enormous AI infrastructure budgets are producing faster cloud and advertising growth. Their response is not to slow construction, but to raise capital-spending plans again.
Together, the four companies are expected to commit roughly $630–$650 billion during 2026. The latest earnings show strong demand and constrained capacity, yet share-price reactions reveal a new concern: revenue is arriving, but the cost of staying ahead is rising just as quickly.
The debate has moved beyond whether AI has customers. It now concerns how long the returns can outrun depreciation, energy and continuous expansion.
Cloud growth validates infrastructure demand
Microsoft reported $82.9 billion in quarterly revenue, up 18%. Azure grew 40% in constant currency, beating expectations, while annualised AI revenue exceeded $37 billion. Microsoft Cloud reached $54.5 billion and commercial remaining obligations rose 99% to $627 billion.
Alphabet’s total revenue grew 20%, its fastest rate since 2022. Google Cloud increased 63%, driven by enterprise AI and infrastructure. Chief executive Sundar Pichai said the company remains compute-constrained, indicating that demand exceeds available capacity.
AWS delivered perhaps the clearest result: revenue rose 28% to $37.59 billion, its fastest growth in 15 quarters. Operating income reached $14.2 billion at a 37.7% margin. Amazon’s custom Trainium and Inferentia chip business passed a $20 billion revenue run rate and is growing at triple digits.
Meta’s return appears through advertising rather than cloud. Revenue grew 33% to $56.31 billion, with its AI-powered Advantage+ system supporting ad performance.
Every major capex forecast moved higher
Microsoft raised fiscal 2026 capital expenditure guidance to $190 billion, compared with approximately $154.6 billion previously expected by analysts. Quarterly spending reached $31.9 billion, up 49%.
Alphabet increased its 2026 range to $180–$190 billion and expects 2027 to rise significantly again. Meta now forecasts $125–$145 billion, citing component prices and additional data-centre costs.
These increases explain why some stocks fell despite operating beats. Investors are not disputing current demand; they are assessing whether future capacity will earn adequate returns after a historic build-out.
Supply constraints are preferable to empty data centres, but they can encourage overbuilding. Equipment purchased for today’s models may face rapid efficiency gains or new architectures before its economic life ends.
The supercycle now faces an efficiency test
Cloud growth of 28–63% supports the infrastructure thesis in the near term. Customers are training models, serving agents and moving more workloads onto accelerated systems. Capacity coming online should unlock revenue that could not previously be served.
The next phase will be judged by utilisation and margin, not installed GPU count. Providers need networking, power and software that keep accelerators busy while routing simpler tasks to cheaper hardware.
Custom chips are strategically important because they can reduce reliance on premium external accelerators and improve economics for common workloads. Amazon’s chip revenue shows that this alternative is reaching meaningful scale.
Energy and construction timelines remain constraints. A data centre cannot earn revenue until power, cooling and networking are available, and local permitting can move more slowly than model demand.
The earnings collectively reject the idea that Big Tech built AI infrastructure without customers. Revenue acceleration is visible across all major clouds, and Meta can point to direct commercial improvement in advertising.
They do not settle whether $650 billion is the right level of spending. That answer depends on demand persisting long enough for capacity to generate returns, and on infrastructure becoming more efficient as it grows.
For now, every company is making the same decision: constrained supply is a bigger risk than excessive capacity. The AI infrastructure supercycle has produced revenue—and that success is encouraging an even larger wager on what comes next.