Google CEO Says Cloud Capacity Is Being Diverted to Frontier AI, Calling 2027 Compute Lockdown 'Extremely Hard'

Google CEO Sundar Pichai, in his first in-depth interview since the launch of the flagship Gemini 4 Argon model, addressed criticism that Google has abandoned frontier AI and confirmed that the company's cloud business is ceding compute capacity to support frontier model development.

Speaking on the Big Technology Podcast, Pichai said the industry is facing a severe compute shortage, with companies currently needing roughly twice the capacity they had planned two years ago. He noted that it is "extremely hard" for customers to secure 2027 compute today and acknowledged that Google Cloud is giving up short-term revenue by prioritizing internal frontier AI research.

On recursive self-improvement (RSI), Pichai said the field is still in a transitional phase where AI agents work alongside human researchers, and remains "quite far" from an autonomous endgame. He argued that frontier model quality directly shapes the ceiling for next-generation models, and cited internal data showing a 30% increase in code sales and a one-third drop in code rollback rates in the four weeks after Argon's launch.

Pichai also pushed back on the idea that frontier AI competition is hitting diminishing returns, saying frontier work offers an informational advantage by surfaceing emerging capabilities early. He also mentioned Google's early-stage project to run TPU test chips in sun-synchronous orbit as part of its long-term compute strategy.