Google is expanding its AI empire — and losing the people who built it

Depending on where you sit, Google either has the most enviable position in artificial intelligence or is bleeding top talent to leading AI labs and other startups on the frontline of innovation.

It’s a contrast that’s been on full display over the past two weeks, beginning with the company reporting 82% revenue growth in its cloud division, followed by a shakeup on Wednesday in Google’s AI organization, as chief scientist Jeff Dean announced his departure after 27 years.

For Google, home to the famous 2017 transformer paper that paved the way for the generative AI boom, the recent events underscore a central challenge facing the $4 trillion company: where to invest. Building frontier models requires huge upfront costs for compute and research with no guarantee of future returns, while the cloud business is proving to be highly efficient and is growing much faster than rival offerings at Amazon and Microsoft.

Alphabet CEO Sundar Pichai said on last month’s earnings call that 90% of Fortune 100 companies are using Gemini Enterprise, underscoring the company’s ability to sell AI services to cloud customers.

Tomasz Tunguz, founder of Theory ventures, said it’s becoming clear that top-of-the-line models aren’t required when it comes to meeting most enterprise demand.

“I think we are at that place with AI, particularly for a lot of white-collar work, where many of the models that are reasonable are good enough,” Tunguz said. “The next evolution of models are likely to be helpful in domains where you have really fancy computers.”

Google’s full-stack approach to AI is a big reason the stock is up 16% this year after jumping 65% in 2025, when it outpaced all of its megacap peers.

It’s been a bumpier road of late. Alphabet shares fell after the latest earnings report due to concerns about capital expenditures, and dipped further on Wednesday following the announcement that Dean is departing and Demis Hassabis is stepping down as CEO of Google DeepMind to become chairman of the unit.

While the tone on Wall Street has been generally favorable, not everyone is celebrating inside of Google.

Some researchers have grown frustrated over access to the computing capacity they need to pursue ambitious projects while watching Google Cloud sell TPUs to outside customers, including Anthropic, according to people familiar with the matter who asked not to be named due to confidentiality. Tensor processing units, or TPUs, are the the company’s homegrown AI chips that compete with Nvidia’s graphics processing units.

Google’s bureaucracy is a common source of frustration, with layers of approval required to move research into products. That can make emerging companies like OpenAI, Anthropic or even younger startups more appealing, especially for AI researchers and developers who prefer lab work to balance sheets.

Dean is leaving along with Google stars Sanjay Ghemawat, Oriol Vinyals, and Quoc Le to start Discovery Loop. On X, Dean said the startup, backed by Google, will be a public benefit corporation “whose mission is to automate machine learning, science, and engineering to accelerate discoveries and progress.”

Their exit follows the departures of other prominent researchers, including Noam Shazeer, one of the authors of the landmark 2017 paper “Attention Is All You Need,” which provided the foundation for generative AI. All eight authors have now left Google. Shazeer left for OpenAI in June, less than two years after Google paid nearly $3 billion to bring him back through an acquihire. His exit came shortly before Nobel laureate John Jumper left DeepMind for Anthropic.

At Google, Dean was one of the very few high-profile voices willing to criticize the Trump administration, and earlier this year he was vocal in opposing the Pentagon’s decision to designate Anthropic as a supply-chain risk, warning that the move could damage the broader U.S. AI industry.

More importantly, from a technical perspective, he helped build the computing infrastructure and neural network systems that established the company as an early leader in modern AI.

Hassabis, who co-founded DeepMind in 2010 and sold it to Google four years later, is becoming chairman of the division and will assume the newly created role of chief scientist at Alphabet, focusing on longer-term research and the societal implications of artificial general intelligence, or AGI. He also plans to devote more time to Isomorphic Labs, the AI drug-discovery company that grew out of DeepMind.

Koray Kavukcuoglu, DeepMind’s technology chief and Alphabet’s chief AI architect, will take over daily management of the division and development of the next Gemini model. According to a person close to the DeepMind team, Kavukcuoglu had been taking on a broader set of responsibilities from Hassabis over the past year, including directing model development and presenting major Gemini releases. Hassabis, meanwhile, has been spending more time away from the lab, focusing on regulation and the longer-term implications of advanced AI.

One of the biggest points of friction inside Google is compute.

Google is investing more than almost any company in the world in data centers, chips and related infrastructure. But capacity remains scarce. Every TPU assigned to training a model, serving a Google product, or fulfilling a contract with a cloud customer reflects a choice among competing priorities.

Frustrations over access to compute can be especially acute when Google announces large infrastructure commitments to competing labs like Anthropic, whose models compete directly with Gemini, sources with knowledge of the matter said.

One of the people said Google has projections for demand in different areas, including research and model training, serving products such as search and Gemini, and working with cloud customers. Those requirements are modeled years in advance, the person said, though capacity may shift over shorter periods if a product grows faster than expected or if priorities change.

Pichai has said on the past two earnings calls that Google continues to prioritize DeepMind’s compute needs even as demand from cloud customers grows. Asked in July about TPU allocation, he said Google’s “first priority” is securing the compute needed to compete at the frontier in AGI development, calling that work “the foundation for everything we do.”

Pichai went on to say that Google balances that need against the capacity required to run its consumer products and AI models, while increasingly placing TPUs directly in third-party data centers to help satisfy external demand.

Dan Niles, founder of Niles Investment Management and a Google shareholder, said access to compute is a natural source of tension.

“Google has all of these other businesses, and they’ve got to figure out who they’re going to give some of these resources to,” Niles said. “Somebody’s always going to be unhappy in that situation.”

In the months that followed, Google ran into a number of snags with its models, most notably delaying the launch of its newest flagship model, Gemini 3.5 Pro. At the same time, Kurian’s business has seen its most explosive growth on record.

Kurian, a former top Oracle executive, has led Google Cloud since 2019, building a bustling enterprise sales organization at a company known for its dominance in consumer internet. Google Cloud designs its own AI chips, operates a global data center network, and sells models, databases, security software, and tools for building AI agents.

The strategy gives Google multiple ways to profit from AI demand. It can sell infrastructure to OpenAI, Anthropic, and other labs, while offering Gemini to enterprises and integrating AI across Search, YouTube, Workspace and its other products.

The emerging question, almost four years into the generative AI craze, is whether Google needs to develop the best AI models or if it’s better off letting other companies foot the bill.

Kavukcuoglu told CNBC at the company’s developer conference in May that Google aims to push the frontier while also improving efficiency. He said the company’s Flash model delivers frontier-level capabilities while running four times faster and more efficiently than comparable models, allowing Google to extend advanced AI across enterprise and consumer services.

Like Tunguz, Niles said the most powerful model is unnecessary for many commercial applications.

“The models are good enough for 90% of what needs to get done,” he said. “You don’t need a Ferrari for this stuff. A Ford will work for 90% of the use cases.”

But for the scientists and researchers trying to produce the next transformer-scale breakthrough, being good enough isn’t always good enough.

WATCH: Demis Hassabis on agentic AI deployment