Google raises 2025 capex to $85B as cloud backlog hits $106B and Gemini reaches 450M MAUs

Summary

Google raised its 2025 capital expenditure guidance to $85 billion from $75 billion, a $10 billion increase driven by overwhelming demand for cloud infrastructure and AI services. The move reflects a sharp shift in the company's willingness to fund lower-margin cloud business at scale.

Google Cloud's backlog sits at $106 billion in unfulfilled demand. The company has that much business it could take if it had the infrastructure to support it. In Q2 2024, Google Cloud missed on topline revenue precisely because it lacked enough GPUs and TPUs to service customers. Yet margins improved because the company had pricing power and refused discounts.

Demand metrics underscore the scale. Gemini reached 450 million monthly active users across mobile, web, and desktop. Google processed 480 trillion tokens monthly across search, AI mode, and API surfaces in May. Daily requests to Gemini grew over 50% from Q1 to Q2.

SSI, Ilya Sutskever's $32 billion AI upstart, will exclusively use Google TPUs for training. That validates Google's infrastructure play at a moment when everyone else is competing for Nvidia GPU capacity. It also suggests SSI believes Google will have surplus capacity to support a massive training run, which aligns with the capex narrative.

Ruth Parat, Google's outgoing CFO, earned praise for cutting waste in the late 2010s. But her $52.5 billion capex spend in 2024 looks too low in retrospect. Ben Thompson argues that Parat may have been too cautious heading into an AI boom, the exact mindset that gets incumbents disrupted. The incoming CFO appears willing to fund AI aggressively.

Paid clicks grew 4% year-on-year, but search revenue grew 12%. That gap means Google is monetizing each click better, not driving more search volume. AI Overviews and other AI-integrated features are monetizing in line with traditional search, according to chief business officer Philip Schindler. The company stopped reporting paid clicks as a core metric, which drew scrutiny from Thompson and others, but the revenue growth speaks for itself.

Google is the only hyperscaler with an at-scale ASIC alternative to Nvidia GPUs in the form of its TPUs. If cloud compute becomes commoditized and competitors are locked into Nvidia, Google's TPU advantage becomes a structural margin benefit. The company effectively captures what would otherwise be Nvidia's margin. OpenAI is heavily dependent on infrastructure partners, now Oracle, while racing to build a consumer product that directly competes with Google Search. That is a much tougher position.

AI is shaping up as a sustaining technology that favors incumbents. Google and Meta both see clear revenue opportunities from AI, both can fund the buildout from profits rather than speculative capital, and both have existing distribution, particularly search.

Google's internal agentic AI projects are slow and expensive today, though the company says it is making progress and has models not yet fully released. Sundar Pichai indicated agentic capability is the highest-priority investment direction. Whether Google ships something to match ChatGPT's agent mode soon will test whether the company's historical caution around risk and deployment speed is an asset or liability.