Google Cloud, AWS, and Azure: Three Speeds, Three Margins, One AI Infrastructure Race

Q2 2026 gives the first clean three-way read on cloud growth — and the numbers tell structurally different stories about scale, margin, and who captures AI-infrastructure economics next.

What Happened

With all three hyperscalers reporting Q2 2026, the cloud infrastructure market produced its cleanest simultaneous read in years — and the topline is acceleration across the board. Google Cloud grew 82% year over year to $24.77 billion, with operating income more than tripling to $8.81 billion and margin expanding from 20.7% to 35.6% in a single year. Its backlog reached $514 billion, the largest forward-demand signal the company has ever reported. AWS grew 37% to $42.2 billion — its fastest growth in 18 quarters — generating $16.6 billion of operating income at a margin near 39%. Both figures come from Amazon’s and Alphabet’s Q2 2026 investor releases.

Microsoft’s Azure grew 43% in constant currency and crossed $100 billion in annualized revenue. That milestone sits between the two rivals on growth rate, but a structural caveat applies here and should frame every comparison that follows: Microsoft does not disclose Azure’s dollar revenue or standalone operating margin — only the growth rate and the annualized milestone. Any revenue or margin ranking that includes Azure is therefore partial, and this piece will not invent a number Microsoft has not published.

The demand signal is unambiguous. Three hyperscalers accelerating simultaneously in the same quarter is not noise — it reflects AI workloads converting to cloud spend at a pace that is broad-based, not concentrated in one vendor’s contract wins. The structural questions are about who captures the economics of that spend at the highest margin, over what capital base, and with which silicon.

The Structural Read

The Map of AI framework — which maps AI economics across nine layers from infrastructure silicon through application monetization — is the right lens here, because the three cloud results are not really a revenue story. They are a story about which layer each company controls, and whether that control is deepening or diluting as AI-native demand scales.

At the infrastructure layer, the decisive variable is silicon — and the CUDA-decoupling story is playing out three distinct ways in these numbers. AWS is pushing Trainium hard: both Anthropic and OpenAI have now committed multi-gigawatt workloads to it, which is the most significant signal in the quarter that a major AI lab is willing to standardize on non-NVIDIA silicon for training at scale. Azure is pairing NVIDIA capacity with its in-house Maia accelerators. Google is leaning on TPUs for Gemini inference and, increasingly, for external customers — and the $514 billion backlog is in part a bet that those customers will accept TPU-native workloads. Whoever’s silicon the frontier labs standardize on shapes the margin structure of the entire AI-cloud stack for the next cycle.

The margin axis is where the structural differentiation becomes legible. AWS at ~39% operating margin is the profit engine of the group — it has scale advantages that Google Cloud is still building toward. But Google Cloud’s margin expansion from 20.7% to 35.6% in a single year is the most structurally significant data point in the set: it suggests the unit economics of AI workloads on TPU infrastructure are improving faster than the market expected, and that Google’s vertical integration (silicon → data center → model → API) is beginning to compound. The Beyond NVIDIA’s Moat thesis is materializing inside these margin lines.

The second axis is capital expenditure, and here the backstop economy framing applies directly. All three are spending at a scale that would have been unthinkable three years ago. AWS deployed $54.2 billion in capex in Q2 alone. Alphabet spent a record $44.9 billion — which pushed its free cash flow to negative $5.9 billion and lifted its full-year 2026 capex guide to $195–205 billion. Microsoft is guiding toward $255–260 billion in capex next year, as detailed in the Azure capex and returns analysis.

The financing divergence matters. Microsoft absorbs that build from a large, diversified, high-margin base — the thesis developed in The Microsoft AI Bet is that Office, Dynamics, and LinkedIn provide the cash-generation buffer that lets Azure spend without market distress. Alphabet’s cloud growth, by contrast, came in the same quarter that its free cash flow went negative — a structural position worth watching if capex remains at this level for multiple quarters. The demand is real; the question is which balance sheet absorbs the build most durably.

Three Implications