Google's Westinghouse-Style Gamble

Google restructured DeepMind's management system on August 5. Demis Hassabis stepped back from day-to-day operations, and Jeff Dean, a key figure in Google engineering, is leaving the company to establish a new lab, Discovery Loop, alongside Sanjay Ghemawat, Quoc Le, and Oriol Vinyals. DeepMind CTO Koray Kavukcuoglu will take charge of operating DeepMind and Gemini.

SemiAnalysis interpreted this as a signal that Google is falling behind in the frontier AI race. The analysis argued that years of conservative compute allocation and a risk-averse organizational culture, along with Google's hesitation to market conversational AI systems it possessed ahead of ChatGPT, have weakened its competitiveness.

However, it is also suggested that Google may have chosen Google Cloud and TPU infrastructure as a larger business opportunity than AI models themselves. SemiAnalysis estimated Gemini's ARR for Q2 2026 at $12 billion, while projecting that GCP's external AI IaaS and TaaS revenue could exceed $73 billion by the end of 2027, with TPU sales potentially reaching $120 billion.

Alphabet's recent quarterly Google Cloud revenue was $24.8 billion, an 82% increase year-over-year. AWS growth was 37%, and Microsoft Azure and other cloud services grew by 43%. SemiAnalysis estimated that TPU system sales added approximately $1.2 billion to Google Cloud revenue in that quarter. However, direct comparisons are limited due to differing scopes of cloud revenue aggregation among companies.

This strategy could prove to be a long-term mistake, similar to IBM withdrawing from the PC business to focus on mainframes and consulting, or Intel reverting to legacy semiconductor businesses from Pat Gelsinger's aggressive investments. Conversely, the article presents the 1880s power competition between Edison and Westinghouse as a new historical analogy, suggesting that Google may be attempting to dominate the AI industry's core infrastructure in a way distinct from the frontier model race.