Google is currently battling it out to keep up with OpenAI and Anthropic in the AI race, and has Chinese labs nipping at its heels, but the field itself was largely created by the company over the last couple of decades.
In a recent interview, Google DeepMind CEO Demis Hassabis made a claim that is hard to dismiss: that roughly 90% of the breakthroughs underpinning the modern AI industry came from Google Brain or DeepMind over the last decade.
“I think a lot of — maybe ninety percent or plus of the big breakthroughs in AI that underpin the modern AI industry came from, you know, Google Brain or DeepMind as we were as separate research entities and now, uh, together as Google DeepMind,” Hassabis said.
The statement is pointed, but it’s not without basis. The transformer architecture — the foundation on which every major large language model today runs, including GPT, Claude, and Gemini itself — was invented at Google. AlphaGo, DeepMind’s landmark reinforcement learning system that beat the world Go champion, helped establish the credibility of deep RL as a serious research direction years before the current wave of AI hype. Hassabis pointed to both as examples of foundational work that competitors have since built their empires on.
This isn’t the first time Hassabis has made his opinions felt on this subject. He’s previously said that Google doesn’t get enough credit for putting transformers out there openly, and that “the whole modern industry is based on that one thing.” Even Google co-founder Sergey Brin has acknowledged that the company underinvested in its own discovery in the years after the 2017 “Attention Is All You Need” paper — essentially handing competitors the keys.
The broader argument Hassabis is making is about research depth. His contention is that Google DeepMind has always run the widest and deepest research bench of any lab, and that this is what separates it from rivals who’ve caught up in terms of commercial products but not fundamental innovation. He said as much when asked whether DeepMind still has the talent to win the race to AGI: “We have by far the biggest and broadest research bench of any of the labs out there, the leading labs out there.”
That claim matters in the current context. The AI talent market has become, in Hassabis’s own words, “ferociously competitive — probably the most ferociously competitive there’s ever been in the tech industry.” Researchers move between OpenAI, Anthropic, Google, Meta, and a growing number of well-funded startups. The idea that any single lab has a permanent lock on talent is getting harder to defend.
Hassabis also used the interview to lay out what Google DeepMind is betting on going forward. Beyond Gemini and its scaling work, the lab is pushing on multimodal generative models — Omni and Veo — which Hassabis sees as crucial for giving AI systems a genuine understanding of the physical world. That grounding, he argued, is going to be necessary for any serious AGI system, and is what will eventually make applications like robotics and smart glasses assistants viable. “To have a full AGI system, you need to be able to also understand the physical world around you,” he said.
It’s worth noting the context here: Hassabis founded DeepMind in 2010, at a time when working on AI in industry was widely considered a dead end. “Definitely not in industry, but even in academia it was basically sort of thought to be career suicide,” he said. The vindication of that bet — and the scale of the breakthroughs that followed — gives him a certain standing to make the 90% claim, even if it’s unverifiable in any precise sense.
Whether or not the number holds up to scrutiny, the underlying point about open research does: the foundations of the current AI industry were largely built in the open, by Google, and many of the companies now racing to build AGI owe their starting position to that. Hassabis isn’t wrong to want that acknowledged.