Alphabet reported $113.8B in Q4 2025 revenue, up 18% year over year. GAAP EPS (earnings per share) was $2.82. For the official filing, see the SEC filing.
Segment check:
Operating income was $35.9B with a 31.6% margin, roughly flat versus last year.
What moved the needle:
Cloud acceleration, real profits: Cloud revenue rose 48% and operating income jumped to $5.3B, a 30.1% margin. This looks less like "growth at any cost" and more like profitable scale. See the SEC filing.
Backlog you can bank on (eventually): Management says Google Cloud's backlog is greater than $240B. In Alphabet language, this means "remaining performance obligations" - contracted revenue not yet recognized. That excludes short-term or cancellable deals. Translation: long-term customer commitments, not cash in hand. See the Financial Times coverage.
Gemini everywhere: Gemini, Alphabet's flagship AI model family, appears across Search, the Gemini app, and enterprise products via Gemini Enterprise and Vertex AI. Alphabet reported selling over 8 million paid enterprise seats in four months - a seat-based subscription metric. Read the Google Blog post.
Jargon, decoded: "Serving unit costs down 78%" means the per-request cost to run model inference fell by 78%. CEO Sundar Pichai attributes this to model optimizations and better infrastructure utilization. Expect more cost drops as scale improves. See the Google Blog post.
Chips, not vibes: TPUs (Tensor Processing Units) are Google-designed AI accelerators that drive much of this efficiency and speed. If you want the technical docs, check Google Cloud TPU docs.
Cost story, not just a victory lap:
GAAP vs "adjusted": Alphabet reported GAAP EPS of $2.82. Some outlets quote adjusted figures, but the official number is GAAP (Generally Accepted Accounting Principles). Operating margin was 31.6%. CFO Anat Ashkenazi noted higher depreciation from data centers and AI hardware as a cost driver - expected after a large AI build-out. See the SEC filing.
AI demand is real and expensive: Capital expenditures are set to rise again in 2026, mostly for technical infrastructure to serve AI workloads. The bill is big and so is the bet. The Financial Times has context.
Why founders should care:
Enterprise AI is table stakes: Eight million Gemini Enterprise seats and a $240B-plus backlog show that buyers expect AI to be built in. Ship AI features or risk being skipped.
Unit economics win: Plan for inference costs early, and then focus on driving them down. Alphabet cut serving costs by 78% - you should model similar gains after optimization.
Tactic: If you build your own stack, budget a 6- to 12-month cost-down sprint focused on inference. If you rent infrastructure, benchmark TPUs versus GPUs for your workload before scaling.
Quick definitions and notes:
Sources: SEC filing, Google Blog post, Financial Times, Google Cloud TPU docs.
Get daily insider tech news delivered to your inbox every weekday morning.
