Particle.news

Gemini 4 Argon Overwhelms Capacity and Forces Google to Rethink Models and Infrastructure

Sundar Pichai says Argon processed about 25 trillion tokens in its first 72 hours, prompting Google to build leaner model variants and expand infrastructure including a Project Suncatcher pilot.

Overview

  • Gemini 4 Argon saw extremely high early demand, with Google reporting roughly 25 trillion tokens processed in the model’s first 72 hours, about three times the start‑up usage of Gemini 3.
  • The surge exposed compute and capacity limits inside Google, and the company is developing more efficient versions of Argon and staging access to expand availability as demand grows.
  • DeepMind has reorganized its leadership with Demis Hassabis stepping back from day‑to‑day duties, Koray Kavukcuoglu taking operational control, and Jeff Dean leaving to start a new venture called Discovery Loop.
  • Google plans to roll persistent consumer agents that run tasks in the background and carry context across Search, Gmail, and Gemini to offer a unified, multi‑product experience.
  • Sundar Pichai defended elevated AI spending tied to customer demand and long‑term contracts and said Project Suncatcher’s first pilot is progressing as a long‑horizon plan to provide constant solar power and room for future data‑center growth.