Google's parent Alphabet announced Q4 2025 earnings February 4, revealing annual revenue above $400B for the first time and setting 2026 capital expenditure guidance at $175-185B. That's roughly double the $91-93B spent in 2025, exceeding analyst expectations of $119.5B.
CFO Anat Ashkenazi broke down the spend: 60% ($105-111B) goes to servers, split evenly between internal workloads and Google Cloud infrastructure. The remaining 40% ($70-74B) funds datacenter construction and networking. The server budget covers both Google's tensor processing units and Nvidia GPUs, serving customers including Apple, OpenAI, and Anthropic.
Google Cloud revenue surged 48% year-over-year to $17.7B in Q4, putting the business at a $70B annual run rate. Cloud backlog rose 55% to $240B. CEO Sundar Pichai noted customers are exceeding their committed spend by 30%, and deals over $1B now surpass the prior three years combined.
What this means in practice: Alphabet is betting heavily that AI infrastructure demand sustains. Pichai flagged power, land, and supply chain constraints as persistent concerns when asked what keeps him up at night. History suggests datacenter capacity buildouts take 18-24 months minimum, creating execution risk if demand shifts.
Worth noting: Shares whipsawed post-earnings, down 3% then recovering, as investors questioned whether the massive spending translates to margin expansion or just higher costs. DA Davidson counters that Cloud's acceleration justifies the investment, particularly given growth now outpaces Azure according to analyst comparisons.
For enterprise teams evaluating Google Cloud, the capex commitment signals capacity availability, but also raises questions about future pricing. The company's push into Vertex AI and enterprise products generated billions in quarterly revenue, according to Ashkenazi, though specific figures remain undisclosed.
The fine print matters here: Google's ad business, which funds this infrastructure bet, grew 13% to $82.3B in Q4. That revenue stability underwrites the AI gamble, but depreciation on fast-cycling AI hardware will pressure margins through 2026.