Sundar Pichai: “We built Transformers to solve product problems” — and now Google is sprinting back into AI

Transformers, LaMDA and the product story

Sundar Pichai told John Collison that Google’s work on Transformers wasn’t ivory-tower research — it was product-driven from day one. He argues BERT and MUM produced some of the biggest jumps in Search quality, and that internal projects like LaMDA were conceived as product experiences long before the wider boom in chat‑style apps. It has been reported that an engineer even believed LaMDA was sentient — a line that landed as equal parts tech lore and corporate cautionary tale during the conversation.

Pichai pushed back on the familiar narrative that Google “invented but didn’t ship.” He said the company used its models inside Search immediately, then iterated. Yet there was regret too: he admitted he wished Google had funded Waymo faster. That small, human moment — a CEO quietly naming a missed acceleration — is the emotional core of the interview. It’s a reminder that even giants stumble on timing.

Bottlenecks, cash and a cultural reset

On the operational side, Pichai framed 2026 as a year of supply constraints: memory and power are the real chokepoints for scaling big models. It has been reported that Alphabet plans to spend roughly $175–180 billion in CapEx in 2026 to fight through those limits, pouring money into chips, data centers — and even long‑shot bets like orbital infrastructure. He’s bullish on AI’s macro impact, saying he expects meaningful U.S. economic growth driven by the technology.

Beyond tech and dollars, Pichai described an internal cultural shift back toward “Googley” optimism: a focus on speed, product rigor, and engineering scale. How Google works, he suggested, is changing in service of one thing — get better models into people’s hands faster. So will that be enough to keep Google on top? For now, the company is betting that capital, chips and a renewed hunger will do the trick.

Sources: cheekypint.substack.com