Everything in Google’s AI stack starts with chips. Google designed TPUs specifically for its own AI workloads, giving it a 2-3 year design iteration advantage over companies dependent on NVIDIA’s allocation cycles.
Ironwood is approaching general availability. Validation is extraordinary: Anthropic is committed to access up to 1 million TPUs, and 9 of the top 10 AI labs now choose Google Cloud infrastructure. This is not a commodity chip business—it’s a vertically integrated compute layer where hardware is co-optimized with JAX and TensorFlow.
The 78% Number
Gemini serving costs dropped 78% in 2025, driven by model optimizations running on proprietary hardware. That single number explains more about Alphabet’s strategic position than any revenue figure. When your foundation gets 78% cheaper in one year while competitors pay market rates for GPU access, every product built on top carries a structural margin advantage.
The Compounding Loop
Invest in custom silicon → 78% cost reduction compounds → every product gets cheaper → more revenue → more chips → repeat. CapEx going from $91B (2025) to $175B+ (2026) isn’t excessive spending. It’s scaling the foundation that makes everything above it cheaper.
The silicon layer creates the cost gravity that pulls margins upward across the entire stack.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
