According to Financial Times News, Google DeepMind chief Demis Hassabis warned that some areas of AI investment look “bubble-like,” specifically pointing to multi-billion dollar seed rounds for startups with no product or technology as unsustainable. He made these comments while discussing Google’s position in the AI race, noting their Gemini app now has 650 million monthly users and features like AI Overview reach two billion. Hassabis stated that despite intense competition, Google had a “really good year” with the release of Gemini 3, which he claims tops the leaderboards. He also announced partnerships with Warby Parker and Gentle Monster to develop new smart glasses, suggesting a universal AI assistant could be the “killer app” this hardware needs. On safety, he emphasized Google’s work on tools like SynthID for watermarking and expressed a need for governmental action for industry-wide coordination.
Bubble vs. substance
Hassabis is drawing a line here, and it’s a crucial one. On one side, you have the foundational tech and infrastructure—the models, the research, the actual user adoption. That’s where Google lives, and from his view, that’s solid. Demand is insane, chips are scarce, and they’re scaling to billions of users. That’s not a bubble; that’s a fundamental shift.
But on the other side? That’s where the froth is. Multi-billion dollar seed rounds for an idea and a PowerPoint? That’s pure speculation. He’s basically saying the money chasing the idea of AI has become detached from the commercial reality of building AI. And he’s right. We’re going to see a brutal correction in that startup layer, especially for companies without a clear path to revenue or a defensible tech moat. The real players, the ones with the models, the compute, and the distribution, will be just fine. The rest? Not so much.
The real competition is multimodal
Here’s the thing I found most interesting. When asked about rivals, he barely mentioned the usual suspect (OpenAI) by name. Instead, he pointed to Anthropic’s work on code with Claude. But then he immediately pivoted to what he’s truly excited about: multimodal AI. This isn’t just a buzzword for him. He’s talking about a model that natively understands text, image, video, and audio because that’s what you need for an assistant that “travels around with you in the real world.”
That’s the real battleground. It’s not just about a better chatbot. It’s about building the perceptual engine for the next generation of devices—smart glasses, robots, you name it. Google’s bet is that their decade-plus of multimodal research (and yes, their history with Glass) gives them a structural advantage. If they can crack that, the chatbot wars start to look like a preliminary skirmish.
The talent war and the mission
The question about researchers getting $100 million offers was telling. Hassabis’s answer was basically, “Sure, we pay well, but that’s not why the best stay.” His argument is about mission and impact. Can you see your work ship to a billion people? Can you work on AI for science, like AlphaFold for protein folding? That’s a powerful retention tool, especially when you’re competing with founders offering lottery-ticket equity.
But let’s be real. It’s also a luxury Google can afford because of its scale. A startup can’t offer that kind of instant, massive deployment. So the talent market is splitting too: mission-driven builders at the giants, and risk-tolerant pioneers chasing huge upside at the startups. Both will pull talent, but for very different reasons.
Safety and the coming techlash
His comments on safety and a potential “techlash” were cautiously diplomatic. He acknowledges the worry is valid, points to Google’s guardrails and watermarking tech, but then admits that industry-wide coordination probably needs government action. That’s a sober assessment. Voluntary guidelines among competitors only go so far.
So what’s the play? He says the industry needs to show “unequivocal benefits” quickly, pointing to AI for science and medicine. It’s a solid strategy. You combat fear and misuse stories with undeniable good-news stories. But the clock is ticking. The pressure for regulation is building faster than those benefits are materializing for the average person. This year will be a tightrope walk between incredible innovation and growing public skepticism.
