AI: Google’s New AI Chief ‘KK’ runs AI Product before Lab. AI-RTZ #1177

As mentioned last week, Google’s AI bench is being re-constituted fast post recent top-level changes. Last week it was three senior changes on the same day: DeepMind Founder/CEO Demis Hassabis stepping up to chairman and chief scientist.

Google veteran employee #30 Jeff Dean and three of Google’s most important builders striking out to found Discovery Loop. And Demis’s #2, Koray Kavukcuoglu (aka ‘KK’), moving from the UK to Google California to head up Google AI. And of course Google Gemini. Reporting directly to Alphabet/Google CEO Sundar Pichai.

Time for a close-up on ‘KK’, who along with Thomas (Tom) Kurian, CEO of Google Cloud, is now the other senior Google exec driving Google AI products and services around the world.

CNBC zeroes in “Google DeepMind: Koray Kavukcuoglu takes over in frontier AI push”:

“DeepMind’s new boss is set to face intense pressure to close the AI performance gap with OpenAI and Anthropic.” (CNBC)

The background: Kavukcuoglu is DeepMind old guard, there since 2012, before Google even acquired it. He was the AI unit’s CTO and Google’s chief AI architect, increasingly directing model development and presenting the major Gemini releases over the past year. Now he steps up to SVP, reporting directly to Sundar Pichai, overseeing Gemini model development, frontier AI research, AND the Gemini app and developer teams. He has moved from London to Mountain View, consolidating Google’s AI leadership at headquarters.

And the consensus external Google complaint he inherits is real, so let me state it plainly.

Google has not shipped a frontier model since early this year. Gemini 3.5 Pro was delayed in July to improve coding performance, and the stock sank on the news. Anthropic’s Mythos/Fable 5 and OpenAI’s GPT 5.6 (and soon Astra), are pulling ahead.

Morningstar’s analyst calls those two ‘miles ahead’ in AI Coding, the first clear killer use case of AI Coding in the Enterprise, which Google is viewed as ‘under-prioritizing’.

And the departures are marquee names: Dean and company to Discovery Loop, Noam Shazeer to OpenAI, John Jumper to Anthropic. Some analysts have gone as far as writing DeepMind out of the frontier race entirely.

Loud negative drumbeats indeed.

My take: the consensus is scoring Google on the wrong game being played by Google.

I have covered Google’s AI ups and downs for years, and the pendulum swings hard both ways: remember OpenAI’s ‘Code Red’ about Google just eight months ago, when Gemini 3 took its victory lap and analysts said it moved the frontier forward. Three things matter more than the frontier AI model-of-the-month scoreboard.

The first is that Kavukcuoglu runs a product, not a lab. Look at his portfolio: models, the Gemini app, and the developer teams under one executive, reporting to the CEO. As one Forrester analyst put it to CNBC, that is how a company organizes a product group rather than a research lab, and it is something Google knows how to do better than OpenAI or Anthropic.

Expect execution over exploration: ship Gemini 3.5 Pro, prove it is not a one-off, hold a release cadence. As I wrote when Hassabis stepped up on AGI in May, the founder was already moving toward science. The formalization tells you what Google wants next: shipping products.

Contrast the people strategies across the AI leaders. Meta’s Zuck BUYS talent: hundreds of billions on researchers and data centers, Alexandr Wang acquired with Scale AI in a $14.3 billion deal.

Elon consolidates his into SpaceXAI across his conglomeration of companies.

OpenAI and Anthropic raid everyone. Google PRODUCES talent. Departures at this level are the price of having a bench worth raiding, and the proof: a company that just produced the founders of Discovery Loop, with Alphabet investing and providing the compute, still fields several thousand people on the Gemini team behind Kavukcuoglu. Do not over-index on a few famous names, as Hassabis’ own biographer put it this week. Nobody actually left the neighborhood. Google is an investor of most of the startups of their departing executives.

The second is efficiency for billions, not benchmarks for AI Researchers. This has been my consistent read: Google is focused on efficient, low-latency models that can scale to billions of users with Gemini, ramping Intelligence Tokens everywhere: Search, Android, Chrome, YouTube, Workspace. Nearly 90% of Fortune 100 companies are using Gemini Enterprise, per Pichai on last month’s earnings call. Google crossed a billion monthly active users of Gemini, ahead of OpenAI’s ChatGPT.

And Gemini is heading inside Apple’s Siri and Apple Intelligence as a paid backend. That is why I have argued Google is the consumer AI company everyone is waiting for, positioned to lead consumer AI along with Apple, besides and with, not against. The monetization of the model, as even the skeptics concede, is doing extremely well.

The third is Google Cloud, the company’s big focus under CEO Thomas Kurian. Google Cloud is the number three cloud behind Amazon AWS and Microsoft Azure, and it is growing like the number one: 82% year over year last quarter at 36% margins, per the numbers Ben Thompson walked through in his Stratechery piece this week.

It is lapping AWS with the AI startups, and the customer list is the AI who’s who: OpenAI renting capacity, Apple’s AI infrastructure, Anthropic on TPUs at massive scale. Inside Google, the allocation fight has tipped the same way: more than a fifth of TPU shipments through late 2027 are reportedly sold directly to Anthropic, and Google’s Q2 led the AI capex ramps to serve exactly this demand.

Google is increasingly allocating the company to its cloud business, up to and including raising equity, with Berkshire in, to fund it. Selling the picks and shovels to your model rivals is not slipping. It is a second business model on the same silicon.

Now the potential glass half empty. The frontier gap is real, the AI Coding gap is real, and the talent outflows are real. If frontier capability turns out to be the whole game, winner take most, the consensus scoreboard will have been the right one, and Google’s ‘plot twists’ will read differently. Both sides of that coin deserve stating.

But my read stays where it has been.

In a world where intelligence gets commoditized, as I argued in yesterday’s piece on Anthropic’s IPO, the winners are distribution and cost structure. Google has both, vertically: its own TPUs, its own models, its own cloud, its own distribution to billions.

The new AI chief’s job is not to win the frontier-of-the-month. Yet.

It is to run the product machine that turns all of that into Gemini for billions, and cloud revenue from everyone else’s AI.

A product, not a lab. That is the right job description for where Google’s new AI Chief sits in this race.

An impact of the AI Tech Wave moving in so many directions. Stay tuned.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)