Alphabet booked $119.80 billion in Q2 consolidated revenue on Wednesday, 24% growth year-over-year, against an LSEG-tracked analyst consensus of $116.93 billion. It’s the 12th consecutive quarter of double-digit growth, and the number that matters most sits inside it: Google Cloud pulled $24.8 billion, up 82% from $13.6 billion a year ago, blowing past the $22.4 billion Street forecast that already implied 64% growth.

That’s the AI capex trade working, on paper.

Operating income landed at $40.77 billion. Net income of $112.1 billion, versus $28.2 billion a year ago, is flattered by roughly $98 billion in net gains, primarily unrealized marks on equity securities. Adjusted EPS came in at $2.85, a hair below the $2.89 LSEG consensus. The stock-securities windfall is the only thing that isn’t structural in the print.

Everything else is. Capex guidance for 2026 now tops out at $205 billion. The cloud backlog sits at $514 billion, a number that puts Alphabet in direct RPO conversation with Microsoft and AWS, and reframes the “Magnificent Seven” AI-capex debate as a question of who’s booking the deferred revenue to justify the spend. Alphabet is.

Sundar Pichai’s supporting statistics were the sort designed for the Bloomberg headline crawl: 22 billion API tokens per minute across Gemini models, 950 million monthly active users on the Gemini App, nearly 90% of the Fortune 100 on Gemini Enterprise, and 2.4 million weekly actives on Antigravity, Google’s AI coding tool.

The awkwardness is one day old. On Tuesday, Google DeepMind shipped Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber, three variants of the small model, while the flagship Gemini Pro remains at its February update. Anthropic has shipped Claude Sonnet 5, OpenAI has GPT-5.6, xAI has Grok 4.5. Google has Flash SKUs and a promise. Pichai called the Gemini 4 pre-training run “a very ambitious effort,” which is the kind of framing executives reach for when the next frontier model isn’t out yet and the current one is aging.

The read: Alphabet is selling the picks-and-shovels story so persuasively that a delayed flagship becomes a rounding error. That worked for Cisco in 1999, right up until it didn’t.

Sources