Alphabet, Microsoft, Meta, and Amazon are collectively spending over $200 billion a year on AI infrastructure and watching their margins compress in real time.

Capital expenditure is money spent on physical assets: data centers, servers, and the specialized chips needed to train and run AI models. It shows up on the balance sheet as an asset, but it hits cash flow immediately.

Alphabet, Microsoft, Meta, and Amazon have guided investors toward a combined annual capex figure that exceeds $200 billion, with the majority earmarked for AI hardware and cloud infrastructure.

Meta has gone further than most, substantially increasing its 2026 capex outlook specifically because of purchases of AI accelerator chips. Those purchases flow through as operating expenses, which means they directly erode reported profits in the near term.

Microsoft and Google Cloud have reported higher operational costs tied directly to AI investments, producing visible margin impacts in their earnings reports.

Analysts at Goldman Sachs and Bernstein have flagged that the projected payback periods for these AI investments stretch into 2027 and 2028. Companies spending aggressively today should not expect to see those investments generate commensurate returns until the second half of this decade, at the earliest.

Margin compression in cloud and AI segments has been a recurring theme in recent earnings cycles. When revenue growth in those segments does not keep pace with the cost of building the underlying infrastructure, the gap becomes a story.

Microsoft’s Azure cloud unit and Google Cloud both report AI-driven revenue growth as a highlight in their earnings presentations. The question investors should be asking is whether that growth rate is accelerating or plateauing, and whether it is happening fast enough to bring the payback timeline closer than the 2027-2028 window that analysts currently project.

Amazon’s calculus runs through AWS, its cloud division, which has leaned aggressively into AI services for enterprise customers. AWS revenue growth is the most direct proxy for whether Amazon’s AI infrastructure spending is translating into commercial demand.

Meta’s AI spending is heavily weighted toward improving the performance of its existing advertising products and building out its own model ecosystem, making the investment harder to evaluate on a quarter-by-quarter basis.

The volatility risk in tech stocks is tied directly to how the market reassesses AI return timelines over the next several earnings cycles. If patience runs out before payback arrives by the 2027-2028 window analysts project, that is when valuations face serious pressure.