AI Giants’ Earnings: The Quiet Signal for Profits

AI giants’ earnings: a quiet signal on the real profitability of the models, critical for equity and thematic ETF investors.

The latest AI giants’ earnings season has just delivered cloud revenue growth still above 20% on average in Q3 2025—but with margins tightening. Between massive CAPEX, energy costs and sometimes-slow monetisation, the AI narrative is finally meeting hard numbers. For investors building a robust asset allocation, the topic has become central: who is really making money on AI, and at what pace?

What the market is really watching: most commentary focuses on AI numbers in press releases – client counts, infrastructure growth, references to “copilots” or assistants. Yet the decisive detail for the next 12 months sits in three lines: CAPEX, cloud operating margin, and energy cost per dollar of AI revenue.

What stands out right now

• Explosive CAPEX: several big tech firms still report annual data center investments above 80–90 billion dollars cumulated in 2025, up around 25% year-on-year → persistent pressure on free cash flow.
• Cloud margins under pressure: cloud segments still post operating margins close to 25–30% at end-2025, but down 1–2 points from the 2023–2024 peak → AI consumes faster than monetisation follows.
• Cautious enterprise customers: generative AI add-on sales remain concentrated in a handful of large accounts, with AI ARPU (average revenue per AI user) growing more slowly than headlines suggest → real adoption, but budget-constrained.
• Heavily loaded thematic ETFs: AI and “future tech” ETFs have absorbed roughly 30–40 billion dollars in net flows over 12 months to end-November 2025 → elevated valuations and heightened sensitivity to earnings disappointments.
• Real rates still positive: with long-end real rates close to 1.5–2% in dollar zones at end-2025, the cost of capital remains markedly higher than in 2015–2020 → not all AI projects will clear the ROI filter.

Decoding the signal: what the news really says

Part of the consensus continues to bet on a straight line: AI growth at +25–30% per year, margins following, and valuations justified. Reading the actual earnings tells a more nuanced story: value first concentrates in the “infrastructure” and “tools” layers, not in every end application.

Observable fact: in 2025, major cloud providers report AI-linked uplift estimated at around 35% of their infrastructure revenue, while many enterprise software vendors report only +8–12% additional growth tied to AI features. This suggests current value capture is happening primarily at the GPU, data center and platform layers, while many software publishers absorb part of the cost to remain competitive. The monetary backdrop to this shift is set out in our breakdown of how central banks set policy and transmit it to markets.

Notable point: in several earnings calls, finance leadership stresses capital allocation discipline more than “all-in on AI.” AI is becoming a line to arbitrate within a new higher-rate cycle. This signals a regime change: a shift from “growth at any cost” to “ROI measurable per use case.” This corporate regime change also reads through the yield curve, as examined in our study of the inverted curve as a regime signal without immediate effect.

This inflection highlights a deeper transformation of corporate economics, where AI no longer falls under marginal innovation but reconfigures cost models, value chains and sectoral competitive advantages – a framework analysed more broadly in the dedicated pillar page on corporates and sectoral dynamics.

Concrete impact: what changes now

For investors, these AI giants’ earnings are no longer just a hype barometer – they act as a sharp filter between structural winners and followers.

• Single-name equities: cloud players with stable or improving margins despite rising AI spend have historically stood out. Cloud operating margins above 28% and a CAPEX-to-revenue ratio below 18% over time have been useful descriptive markers of operational discipline.
• AI ETFs: AI thematic ETF exposure within diversified equity portfolios has historically remained in low double-digit shares for dynamic profiles, framed within a barbell strategy logic (diversified core plus a targeted AI sleeve). Beyond such levels, sectoral concentration risk has been documented.
• “User” corporates: at the corporate level, AI projects that reduce costs (customer support, back-office automation) have historically delivered more measurable margin impact at 18–24 months than purely “image” projects. The issue is not the technology, but the margin gain.
• Risk management: partial profit-taking after each strong earnings sequence, with reallocation toward more defensive sectors, has been documented as a recurring practice – as discussed within the 50-30-20 rule framework.

Counterargument worth keeping in mind: if AI-driven productivity accelerates strongly from 2026, profitability could catch up to current investment levels much faster than expected, and a market positioned cautiously could find itself underexposed to pure software players.

Micro-trends that matter

• Energy cost per AI query: several groups report a 15–20% rise in their energy bill linked to AI workloads in 2025. Tracking the energy cost / AI revenue ratio – even approximately – is becoming a decisive KPI.
• Mix of recurring AI revenue: per-user AI subscriptions, rather than per-query credit, raise revenue visibility. A share of “recurring AI revenue” above 60% is a quality signal.
• AI customer churn: early figures on AI cancellations show rates sometimes above 8–10% annually on offerings not embedded in core workflows. If this persists, certain business models will need to be reworked.
• AI’s share of total CAPEX: when AI exceeds 50–60% of a group’s CAPEX, sector overinvestment risk rises. This indicator is still under-watched but can precede rationalisation plans.

3–12 month outlook

Scenario 1 (central): orderly digestion — AI giants maintain cloud revenue growth above 20% in 2026, margins stabilise around 27–29%, and the market accepts a tighter cash flow regime. Valuations remain elevated, but without a fresh massive multiple expansion. In the same vein: The trillion-dollar build-out in context.

Scenario 2: positive productivity surprise — Customer corporates begin documenting productivity gains above 5% on selected functions from 2026. AI budgets normalise upward, benefiting specialised software publishers most. The market re-rates the software “moat.” See also: the decomposition of how exchange rates track rate regimes.

Scenario 3: abrupt reallocation — If AI growth falls back to around 15% with CAPEX still elevated, rotation toward more cyclical or value sectors becomes likely, in line with signals already observed in certain sectoral rotations in 2025. The most expensive names without a clear margin trajectory would be the most exposed.

In every case, the key remains discipline: tracking the “CAPEX + margin + monetisation” trajectory quarter after quarter, rather than focusing on the technological narrative alone.

Three takeaways

• AI giants’ earnings reveal above all a tight arbitrage between massive CAPEX and cloud margin, more than unlimited cost-free growth.
• For an equity portfolio, AI thematic exposure at low double-digit share, articulated with a diversified base, has historically limited concentration risk.
• The KPI to watch: the trajectory of cloud AI segments’ operating margin each quarter, a concrete barometer of real value creation.

We will revisit tomorrow with a possibly different market.

Last updated — 12 July 2026