In the tech world, there is a massive difference between discovering fire and being the one who sells the first lighter. Demis Hassabis recently pointed out that while OpenAI’s Sam Altman is the face of the AI revolution, the actual engine under the hood of ChatGPT was built in Google’s labs. It’s a bit like Google inventing the wheel, only to watch OpenAI drive off in a Ferrari.
Every major AI you use today, from ChatGPT to Gemini, relies on a technology called the Transformer. In 2017, a group of Google researchers published a paper called “Attention Is All You Need.” It introduced the Transformer architecture. But the irony behind it is that Google had this world-changing tech sitting on its shelf for five years. They used it to improve Search slightly, but they didn’t turn it into a world-shaking product. Then, OpenAI saw that same paper, realized its potential, and used it to build the “T” in GPT (Generative Pre-trained Transformer).
Why Did Google Wait? Hassabis admits that Google may have been “a little bit slow” to scale the tech. But his reasoning is human: Responsibility. When you have billions of users, you can’t just release a “hallucinating” chatbot that gives wrong medical advice or weird political takes. Startups like OpenAI had the privilege of moving fast and breaking things, while Google was stuck trying to ensure things didn’t break.
Hassabis isn’t just looking back with regret. He says the last two years have forced Google to find its “scrappy roots” again. DeepMind and Google Brain were merged into one giant “engine room” to catch up. They are now shipping updates at a ferocious pace—moving from being a quiet research lab to a competitive product machine.
The battle in 2026 isn’t about who wrote the first research paper anymore; it’s about who builds the tool people can’t live without. Hassabis’s comments are a rare admission that Google’s biggest mistake wasn’t a lack of genius, it was a lack of speed. They invented the future, they just forgot to invite everyone to it until someone else did.
