Why Google Turns Down Revenue to Build Better AI

Sundar Pichai, CEO of Google and Alphabet · Big Technology Podcast

Sundar Pichai is the CEO of Google and its parent company, Alphabet. In 2015 he declared Google an "AI first" company. Google recently launched Gemini 4 Argon, a frontier model that beats OpenAI's and Anthropic's public models on several benchmarks in coding, math, science and knowledge work.

Summary

Pichai argues that staying at the frontier of AI is non-negotiable for Google, so frontier research gets first claim on compute even when Google Cloud gives up short-term revenue. Compute is planned 2 to 3 years ahead, which is why everyone is short today; he doesn't see a dark-fiber bubble because Google uses everything it has. On safety, he describes defense in depth, from alignment research to real-time monitoring, and calls for independent evaluators. Product-wise, Google is building persistent agents that run in the cloud and follow users across Gemini, Search and Gmail. These signals are from an interview that Alex Kantrowitz did with Sundar Pichai on Big Technology Podcast.

1. Back at the Frontier. The conventional wisdom had been that Google settled for fast, good-enough models. Pichai calls that "very wrong." Google took longer with Argon than with past models to test and deploy it carefully given its capabilities. In its first 72 hours, Argon processed almost 25 trillion tokens, about 3 times what Gemini 3.0 did, and Google is capacity-constrained serving it. He still sees pockets, including parts of coding, where it needs to improve. ▶ 1:24
   
   Quote: “I think we are so committed to being at the frontier. I think it's in our DNA. We wouldn't do it any other way.” (Sundar Pichai, ▶ 1:47)

2. Why the Frontier Matters. Pichai rejects the argument that standard models are good enough for Google's products. Being at the frontier gives an earlier look at emerging capabilities, some cloud customers want the best model, and frontier models will be better at developing the next models. Deploying Argon to Google's own developers produced a steep inflection, solving problems earlier models couldn't. He agrees good products can run on smaller, cheaper models, which is why Google ships Flash and Flash Lite too. ▶ 4:52
   
   
3. Self-Improvement, Today. Sergey Brin has said that relying more on the model to help develop the model was a core innovation behind Argon. Pichai describes today's state as recursive self-improvement loops alongside human researchers: agents helping researchers in partly automated ways. He says the industry is "pretty far" from the end state of fully self-improving AI. He expects progress along that continuum into next year. ▶ 9:39