Hyperscalers are on pace to spend $660 billion to $690 billion on AI infrastructure by 2026, nearly double 2025's outlay, and the gap between capital deployed and revenue earned is becoming impossible to ignore. Nvidia's dependence on a handful of giant customers exposes it to that gap, while Microsoft's dual role as builder and seller of AI services gives it more ways to offset the risk.
The spending spree in numbers
Big Tech's AI buildout is now the most expensive infrastructure project in corporate history. Hyperscalers spent roughly $380 billion on AI-related capital expenditures in 2025. That figure is projected to reach $660 billion to $690 billion by 2026.
Microsoft is leading the charge, guiding toward AI-related expenditures exceeding $120 billion in fiscal 2026. Its most recent fiscal third quarter saw spending between $34.9 billion and $37.5 billion, a 75% year-over-year increase. Aggregate hyperscaler AI infrastructure commitments from 2025 through 2027 are projected to approach or exceed $1 trillion.
Nvidia's dominance carries three risks
Nvidia captures an estimated 90% of AI accelerator spending. At current buildout rates, that translates to approximately $180 billion in annual GPU purchases. But that dominance concentrates its revenue among five or six customers making synchronized bets on a technology whose payoff timeline remains uncertain.
Microsoft, Google, and Amazon have also invested heavily in designing their own AI chips, and even a shift of 10 to 15 percentage points in market share would reshape a market worth hundreds of billions. On top of that, analysts describe what they call the "trillion-dollar capex gap": current AI revenues across the industry remain in the low tens of billions annually, far short of what the pace of investment needs to produce by 2030.
The company sits on both sides of the trade
Microsoft sits on both sides of the AI value chain — it builds the infrastructure and sells the services layered on top. Azure's AI offerings, Copilot integrations across Office, and enterprise AI tooling give it several direct paths to monetize its own capital spending.
That dynamic showed up in Microsoft's latest results: revenue growth exceeded $37 billion. Its Azure order backlog of $80 billion, even if constrained by power availability, provides revenue visibility most tech companies lack. The company's data-center lease commitments now exceed $300 billion, a bet that AI compute becomes as fundamental to enterprise operations as cloud computing did over the past decade.
For Nvidia, the metric to watch going forward is not revenue growth but revenue concentration and the pace of custom-silicon adoption among its biggest customers.
Source: Crypto Briefing
Trading involves risk.
Risk Disclaimer
