Big Tech capex for AI infrastructure jumps as investors balk at 2026 spend

Big Tech is signaling that the next competitive moat in AI will be owned infrastructure: data centers, custom chips, and the physical supply chain needed to run models at scale. The numbers for 2026 are now coming into view—and the immediate investor response has been negative.

2026 capex projections concentrate around compute and hardware

Amazon is guiding to 200 billion dollars in capital expenditures through 2026 spanning “AI, chips, robotics, and low earth orbit satellites,” up from 131.8 billion dollars in 2025, per its earnings release (source). For marketers and commerce operators, the practical implication is that AWS will keep pushing supply-side advantage: more capacity, more specialized silicon, and potentially better unit economics on inference over time.

Alphabet is also stepping on the gas, projecting 175 billion to 185 billion dollars of capex for 2026 versus 91.4 billion dollars the year prior, according to its earnings release (source). Meta has separately projected 115 billion to 135 billion dollars in 2026 capex, while Oracle is cited at about 50 billion dollars (source). Microsoft has not issued a full-year 2026 figure, but a recent quarterly capex number of 37.5 billion dollars implies roughly 150 billion dollars annualized if sustained (source).

Investor pushback raises pressure to justify ROI beyond “more compute”

Despite the industry’s internal logic—compute becomes the scarce input, so own as much as possible—stocks reportedly fell as investors reacted to the scale of commitments. For B2B marketers, this matters because pricing and availability of model APIs and managed AI services are downstream of capex cycles.

Expect more messaging that reframes spend as efficiency (custom chips, robotics, data center utilization) rather than an arms race, alongside sharper proof points on monetization—especially from vendors selling AI features into cloud, ads, and enterprise SaaS.