Artificial intelligence returns on investment across hyperscalers Report Interpretation

Overview

This Citi Research Viewpoint examines AI returns across US hyperscalers. Its central conclusion is that accelerating AI capital expenditure, persistent capacity constraints and rapidly expanding enterprise usage are driving revenue and profitability more strongly than consensus currently assumes.

Core views

Citi's central argument is that the economics of AI investment remain attractive despite the scale of infrastructure spending. Hyperscaler cloud revenue growth accelerated to 48% in 2Q from 39% in 1Q, while collective backlog growth accelerated to 151% year on year from 143% in the prior quarter. Citi interprets these trends as evidence of continuing enterprise AI adoption. Although trailing-12-month cash returns on cash invested (CROCI) were broadly stable at 28%, versus 29% in 1Q, the report argues that current returns support continued aggressive AI infrastructure investment. Its conclusion is summarized as “CapEx + Time = Profits”: the spending acceleration beginning in 2024 should feed through to later revenue and profitability. The report points to demand and monetization indicators across the major cloud platforms. Nearly 500 Google Cloud Platform customers consumed 1 trillion tokens, more than 2,000 enterprises consumed over 100 billion tokens over the past year, Microsoft had 30 million paid Microsoft 365 Copilot seats, and Amazon reported an AI revenue run rate above $25B. Higher compute prices amid strong AI demand are supporting revenue, backlog and margins: hyperscaler EBITDA margins held at 43% in 2Q, unchanged from 1Q. Citi therefore views increasingly negative free cash flow and related financing-cost concerns as insufficient reasons for hyperscalers to curb investment while returns remain high. Capital-spending guidance reinforces the demand thesis. The aggregate CY26E CapEx forecast is now $730B, up 77% versus CY25, after consensus raised its combined 2026 CapEx forecasts by 0.8% following results. Meta increased guidance to $130B-$145B from $125B-$135B. Google raised its range by 7% at the midpoint to $190B-$205B, implying 116% year-on-year growth versus consensus at 105%, citing faster capacity delivery to meet demand. Amazon's outlook rose by $20B to about $220B, or 67% year-on-year and 10 percentage points above consensus, because of higher memory costs. Google expects a significant increase in 2027 CapEx versus 2026, Microsoft expects FY27 CapEx growth year on year, and Amazon expects capacity constraints to persist into 2027, with substantial demand already in its 2028 pipeline. Citi links prior CapEx growth to an accelerating cloud-revenue curve using revenue growth overlaid on CapEx growth pulled forward by 36 months. In 2Q, GCP growth reached 82% year on year, a nearly 20-point sequential acceleration and 17 points ahead of consensus. AWS grew 37%, 6 points ahead of consensus, and Azure grew 43% in constant currency, 3 points ahead. Google, Microsoft and Amazon's combined backlog expanded to $1.69T from $1.45T in 1Q. While consensus revised 2027 revenue forecasts up 2.3%—the first time Citi observed revenue revisions exceeding CapEx revisions—the institution believes this still understates the growth curve. It estimates consensus is underestimating revenue growth by 500bps for 2027 and 900bps for 2028, and expects this dynamic to support outperformance across AI infrastructure, compute, applications and AI-enabled end users.

Analysis framework

Citi combines hyperscaler earnings results, management CapEx guidance, cloud revenue and backlog data, enterprise AI-usage indicators, margins and CROCI. It also compares trailing-12-month revenue growth with CapEx growth shifted forward by 36 months to assess how prior investment may translate into future revenue growth.

Methodology notes

• Corporate Fundamentals and FinanceFree cash flow analysis

  Cash Returns on Cash Invested (CROCI) and free-cash-flow consequences of AI investment

  The report uses CROCI to judge whether high AI infrastructure spending is producing acceptable cash returns, while noting that free cash flow is becoming more negative as investment rises.

• Industry AnalysisSupply-demand framework

  AI compute demand relative to infrastructure capacity

  Citi uses capacity constraints, backlog expansion, customer usage and compute pricing to argue that demand continues to exceed supply and supports spending and revenue growth.

• Other

  Revenue growth overlaid on CapEx growth pulled forward by 36 months

  Citi compares past CapEx growth with later trailing-12-month revenue growth to support its view that recent investment should drive a stronger revenue curve from 2027 onward.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Alphabet (GOOGL)

  Hyperscaler benefiting from accelerating GCP demand and higher AI capacity investment.

  Strengths

  GCP grew 82% year on year in 2Q, 17 percentage points ahead of consensus; Google raised 2026 CapEx guidance to $190B-$205B.

  Comparison

  GCP showed the strongest reported cloud-growth acceleration among the cited platforms.

  Risks

  Capacity demand continues to outpace investment.

• Meta Platforms (META)

  Hyperscaler increasing AI infrastructure investment.

  Strengths

  Meta revised its CapEx guidance to $130B-$145B from $125B-$135B.

• Amazon

  Hyperscaler with AI-driven AWS growth and elevated infrastructure spending.

  Strengths

  AWS grew 37% year on year in 2Q, 6 percentage points above consensus; Amazon cited an AI revenue run rate above $25B.

  Weaknesses

  Higher memory costs contributed to a higher CapEx outlook.

  Comparison

  AWS growth trailed GCP but exceeded consensus expectations.

  Risks

  Capacity constraints are expected to persist into 2027.

• Microsoft

  Hyperscaler with AI demand supporting Azure growth and continued CapEx expansion.

  Strengths

  Azure grew 43% in constant currency in 2Q; Microsoft had 30 million paid M365 Copilot seats.

  Comparison

  Azure beat consensus growth expectations by 3 percentage points.

  Risks

  Customer demand continues to exceed supply.

Key data

• Hyperscaler cloud revenue growth+48% in 2QAccelerated from +39% in 1Q.
• Collective backlog growth+151% year on year in 2QAccelerated from +143% in the prior quarter.
• CROCI28% in 2QEssentially flat versus 29% in 1Q.
• Hyperscaler EBITDA margin43% in 2QUnchanged from 1Q despite rising AI CapEx.
• Combined Google, Microsoft and Amazon backlog$1.69T in 2QUp from $1.45T in 1Q.
• Aggregate CY26E CapEx$730B77% above CY25.
• GCP revenue growth82% year on year in 2Q17 percentage points ahead of consensus.
• Consensus revenue-growth understatement+500bps for 2027 and +900bps for 2028Citi estimate for AI-related revenue growth.

Impact & implications

Citi expects sustained AI investment to benefit infrastructure, compute, applications and AI-enabled end users. It argues that the combination of stronger cloud growth, backlog accumulation, high returns and persistent supply constraints makes consensus revenue and profitability expectations too low from 2027 onward.