Goldman Sachs projects AI capital expenditures to reach $1.2T by 2027

Goldman Sachs has ratcheted up its forecast for how much America’s biggest tech companies will spend building out AI infrastructure, projecting US hyperscaler capital expenditures could climb as high as $1.4 trillion by 2027. That figure blows past the bank’s own prior estimates of roughly $1.1 trillion and dwarfs a broader Wall Street consensus that had been hovering closer to $920 billion.

The numbers behind the ramp

Goldman’s updated trajectory sketches out a steep climb. The bank previously estimated hyperscaler AI capex at roughly $405 billion in 2025, rising to approximately $750 billion in 2026, before hitting the $1.2 trillion range in 2027. The latest revision pushes that 2027 figure even higher.

The companies driving this wave are the usual suspects: Microsoft, Amazon, Alphabet, Meta, and Oracle, with OpenAI also drawing notable investment activity. These firms are collectively transitioning from what Goldman characterizes as an experimental phase of AI deployment into full-scale commercial implementation.

Looking further out, the bank estimates cumulative AI infrastructure spending could reach approximately $7.6 trillion from 2026 through 2031. That breaks down into roughly $5.1 trillion for compute, $2.1 trillion for data centers, and $358 billion for power infrastructure.

Goldman’s analysts have drawn comparisons to historical technology waves, specifically the buildout of railroads and automobiles. Both required massive upfront capital deployment before the downstream economic benefits materialized.

Debt is doing the heavy lifting

More than one-third of the projected 2027 capital expenditure, roughly $400 billion, is expected to come through investment-grade bond issuance. That’s a meaningful shift from the self-funded, cash-flow-driven investment model that defined big tech’s first two decades.

Goldman points to advertising and subscription models as the primary monetization channels for AI-driven tools. The bet is that consumer AI agents and enterprise AI applications will generate revenue streams large enough to justify the infrastructure investment.

Valuation tensions and volatility risks

Goldman notes that median AI infrastructure stock valuations sit at around 26 times forward price-to-earnings ratios.

The bank’s analysts also flag several supply-side constraints that could slow the buildout regardless of how much money hyperscalers are willing to throw at it. Power availability is near the top of the list. Land accessibility presents a similar bottleneck. Memory chip affordability rounds out Goldman’s list of constraints, with AI workloads being extraordinarily memory-intensive.

The shift toward debt financing also creates a secondary market dynamic worth monitoring. Over $400 billion in new investment-grade bond supply will need to be absorbed by fixed-income markets.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.