Pichai interview: compute prioritized for Gemini 4 Argon

According to InfoQ Chinese's write-up of a Big Technology Podcast interview, Google CEO Sundar Pichai spoke with host Alex Kantrowitz about Google's investments and trade-offs in the AI race. Pichai discussed the recently launched Gemini 4 Argon, whose usage reached close to 25 trillion tokens within its first 72 hours — roughly triple Gemini 3.0's figure over the same period — while compute supply fell short of demand. Responding to the outside perception that Google might abandon frontier AI competition and focus instead on folding adequately capable models into its existing products, Pichai pushed back hard, calling that judgment wildly wrong. He revealed that to protect frontier model research, Google has prioritized compute allocation to its AI research teams, a decision that has cost the cloud business some short-term revenue.

On the internal development front, Pichai said AI has become deeply embedded in Google's code workflows. As recently as last year, he had stated that 25% of Google's code was AI-written, but when asked the question again he changed his answer, saying that figure no longer carries much meaning. The metrics he offered instead: in the four weeks after Gemini 4 Argon shipped, Google's code change volume rose about 30%, rollbacks dropped by roughly one third, and developers began using Gemini 4 Argon on complex projects spanning millions of lines of code. Rather than the share of AI-generated code, Google now focuses on indicators such as code change volume and rollback behavior, which Pichai argued better reflect AI's impact on development efficiency and quality. Pichai also touched on topics including AI's role in model development and continuously running personal agents. So far, only fragments of those parts of the conversation have been reported, and the full interview has not yet been published.