Google's new AI chief has one job: close the gap with OpenAI and Anthropic
Demis Hassabis steps aside as DeepMind boss. His replacement, Koray Kavukcuoglu, takes the wheel at a company that analysts say has fallen behind in the race for the most powerful AI models.
Key points
- Koray Kavukcuoglu became Google DeepMind's senior vice president in mid-2026, replacing co-founder Demis Hassabis, who moves to a chair role.
- Google has not released a frontier AI model (one that leads the industry on performance benchmarks) since Gemini 3.1 Pro in February 2026.
- Rivals Anthropic and OpenAI released well-received new models in 2026, with analysts saying Google fell behind partly by under-investing in coding capabilities.
- Alphabet shares fell when the leadership change was announced, though the stock is still up roughly 76% over the previous 12 months.
- Nearly 90% of Fortune 100 companies use Gemini Enterprise, Google's business AI product, showing strong commercial traction despite the model-performance gap.
Google has a new person in charge of its most important technology. Koray Kavukcuoglu, a DeepMind veteran who joined the lab back in 2012, before Google even acquired it, has taken over as the head of Google DeepMind. His predecessor, Demis Hassabis, the scientist who built DeepMind into a world-famous AI research organisation, moves to a chair role.
The timing matters. Kavukcuoglu is stepping into a company that has, by most analysts' accounts, let its lead slip.
How far behind is Google, really?
Gemini 3, released in November 2025, genuinely impressed analysts. It moved the goalposts. But 2026 has been quieter on Google's side while rivals moved fast. Anthropic, the company behind the Claude family of AI assistants, released a model called Mythos. OpenAI followed with GPT-5.6. Both got strong reviews.
Google's most recent frontier release was Gemini 3.1 Pro, in February 2026. Since then, nothing that analysts would call a step-change.
The specific area where Google has fallen furthest behind, according to Morningstar senior equity analyst Malik Ahmed Khan, is coding. AI assistants that can write and fix computer code have become the dominant use case for businesses buying AI tools, and Anthropic and OpenAI are, in Khan's words, "miles ahead" on that front. Google, he told CNBC, was distracted by chasing revenue in search and other areas and missed the moment.
Google set up an internal team called Code Strike to work on exactly this problem. Whether it is enough, and fast enough, is what Kavukcuoglu now has to prove.
What changes under the new boss?
Kavukcuoglu reports directly to Alphabet CEO Sundar Pichai and now oversees Gemini model development, frontier AI research, and the Gemini app and developer teams, all under one roof. That last bit is significant. Brian Hopkins, an analyst at Forrester, told CNBC this structure looks less like a research lab and more like a product team, something Google knows how to run.
Analysts expect faster, more predictable model releases. The most immediate test is Gemini 3.5 Pro, a model that was reportedly delayed after Alphabet's share price dropped in July when Bloomberg broke the news of the delay.
"The first step would be to ship Gemini 3.5 Pro if possible, and then prove it wasn't a one-off by maintaining a predictable release cadence," said Nick Patience, AI lead at the Futurum Group.
There is also a quieter geographic shift. Hassabis ran DeepMind from London, which helped build the UK's AI ecosystem. Kavukcuoglu already moved to Google's Mountain View, California headquarters. A Google spokesperson said the company remains committed to its London presence.
Should Google users be worried?
Not urgently. The models powering Google's consumer products, including the Gemini app on Android phones, are still competitive for everyday tasks. Google's business products are doing well commercially: Pichai said on a recent earnings call that nearly 90% of Fortune 100 companies use Gemini Enterprise.
As Sebastian Mallaby, author of a Hassabis biography, put it plainly: the monetisation is working even if the model ranking is not yet where Google wants it. The Gemini team, he noted, numbers several thousand people. This is a course correction, not a collapse.
For ordinary users, the most visible near-term change will be whether Gemini gets meaningfully better at helping with technical tasks like writing code or building spreadsheet formulas. That is the benchmark Kavukcuoglu's tenure will be judged against.
Common questions
What is a frontier AI model?
A frontier model is one that sits at the very top of performance rankings across tasks like reasoning, coding and science. Think of it as the AI equivalent of a Formula 1 car: not what most people drive, but the technology that eventually trickles down into everyday products.
Does this affect the free version of Gemini I already use?
Not immediately. Leadership changes and research priorities take months to show up in consumer products. The free Gemini assistant keeps working as normal while Google works on its next-generation models behind the scenes.
What happened to Demis Hassabis?
Hassabis, who co-founded DeepMind and led it for over a decade, moves to a chair role. He remains involved with the organisation but steps back from day-to-day leadership.
