What Happened

Goldman Sachs has projected that AI hyperscaler capital expenditures could reach $700 billion in 2026, a figure that would match peak spending levels during the late 1990s telecom boom and represent a potential $200 billion upside from current consensus estimates. The projection underscores an unprecedented infrastructure buildout by Big Tech giants racing to secure their positions in the artificial intelligence economy.

The forecast focuses on major hyperscalers—Microsoft, Amazon (AWS), Google (Alphabet), and Meta—whose combined AI-related data center investments are driving the surge. According to Goldman's analysis, this spending wave would bring total hyperscaler capex from 2025–2027 to approximately $1.15 trillion, more than double the $477 billion spent from 2022–2024.

By the Numbers

The scale of investment is staggering by any historical measure:

  • $700 billion — Goldman's projected 2026 AI hyperscaler capex, representing 1.5% of GDP and far above current spending at 0.8% of GDP [Goldman Sachs]
  • $527–667 billion — Range of current consensus estimates, which Goldman suggests have consistently underestimated actual spending momentum [Investing.com]
  • $106 billion — Q3 2025 hyperscaler capex spending, up 75% year-over-year [Introl]
  • 75% — Portion of 2026 capex (approximately $450 billion) targeting AI infrastructure specifically, including GPUs, servers, and data centers [MUFG Americas]
  • $360 billion — Palantir's market capitalization following the Pentagon's formalization of its Maven AI system as a permanent military program [Channel News Asia]

Individual company projections show the magnitude of commitment: Amazon, Microsoft, Google, and Meta each plan to exceed $100 billion in capex for 2026, with Amazon leading at over $125 billion, primarily for AWS AI infrastructure.

Market Reaction

Investors have rewarded the infrastructure buildout theme handsomely. AI infrastructure stocks returned approximately 44% year-to-date as of late 2025, with semiconductor names capturing significant gains. Nvidia, which commands an estimated 85% market share in AI accelerators, saw its stock rise 34% in 2025, though AMD outperformed with an 82% gain on competitive positioning in the AI chip market.

The market has also recognized emerging AI infrastructure players beyond the hyperscalers themselves. Palantir's stock doubled over the past year following the Pentagon's decision to formalize its Maven AI system as a "program of record"—providing stable, long-term funding and streamlined adoption across all military branches. The Maven contract ceiling was increased to $1.3 billion through 2029, signaling sustained government commitment to AI infrastructure.

However, Goldman anticipates capex growth rates will peak in late 2026 before decelerating, potentially shifting investor focus from spending announcements to earnings realization and return on investment metrics.

The Bigger Picture

The investment wave raises fundamental questions about the economics of AI infrastructure. While productivity gains from AI adoption are reportedly widespread—with 96% of organizations experiencing some productivity improvements according to an EY survey—the translation into measurable financial returns remains inconsistent.

Research presents a nuanced picture: while three-quarters of business leaders report positive returns on AI investments according to a Wharton survey, a BCG analysis found that 60% of companies are capturing hardly any material value from AI despite substantial investment. The gap appears to stem from implementation maturity—organizations investing $10 million or more across multiple business units are significantly more likely to see substantial gains.

The infrastructure spending also creates downstream economic effects. Goldman research indicates the AI buildout is contributing to higher electricity costs for consumers, as data centers demand unprecedented power capacity. The firm projects hyperscalers will need to raise approximately $1.5 trillion in total funding through cash flows and debt to finance the expansion.

What to Watch

Several indicators will determine whether this infrastructure bet pays off:

  • Capex trajectory — Watch for whether 2026 spending reaches Goldman's $700 billion projection or tracks closer to the $527–667 billion consensus range
  • ROI realization — Monitor enterprise AI adoption surveys for shifts in measured productivity gains and financial returns
  • Government contracts — Track Pentagon and federal AI procurement following Palantir's Maven formalization, which could set precedents for broader public-sector AI infrastructure spending
  • Infrastructure bottlenecks — Power availability, semiconductor supply, and data center construction capacity may constrain the buildout timeline
  • Debt markets — The $108 billion in debt raised by hyperscalers in 2025 suggests financing is available, but rising rates could impact the economics of leveraged infrastructure investment

The parallels to the late 1990s telecom boom are instructive: massive infrastructure investment eventually delivered transformative capabilities, but not before a period of overcapacity and consolidation tested investor patience. For Big Tech's AI infrastructure bet, the payoff timeline remains the critical unknown.