STRAIGHT TO THE POINT
Over the span of just two weeks, Google announced 82% growth in its cloud revenue. Then, almost immediately, the company's chief scientist, Jeff Dean, announced he was leaving after 27 years.
The stock fell more than 3% that same day.
And that is what makes the story interesting. Because these two headlines seem to tell two completely different stories about the same company.
WHAT EXACTLY HAPPENED
A lot happened at once.
First, Jeff Dean is leaving Google. He joined the company in 1999, just one year after it was founded. He co-founded Google Brain around fifteen years ago and helped develop Google's TPU chips, the custom processors that power much of its AI infrastructure.
In short, he is one of the people who literally helped build the foundation on which Google's AI business runs today.
And he is not leaving alone.
He is joined by Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. Together they are launching a new company called Discovery Loop, whose mission, according to Dean himself, is to automate machine learning, science, and engineering in order to accelerate discovery.
The first investor in Discovery Loop is Google.
Its cloud provider will also be Google.
Second, Demis Hassabis is stepping down as CEO of DeepMind. He will become President of the AI division and Chief Scientist of Alphabet. His focus will shift toward the bigger picture, while also dedicating more time to Isomorphic Labs, the AI-driven drug discovery company.
"I've worked toward AGI my entire life, and now it feels close," he reportedly told colleagues.
His successor is Koray Kavukcuoglu, DeepMind's former CTO. A thirteen-year veteran of the company, he helped establish its deep learning efforts and worked on projects such as WaveNet and DQN.
He will report directly to Sundar Pichai and oversee Gemini models, research, and developer teams.
THE TALENT CYCLE IS NOTHING NEW
This is where it gets interesting.
Many people are talking about a talent exodus as if it began yesterday.
It didn't.
Remember the famous 2017 paper, "Attention Is All You Need"? The paper that paved the way for today's generative AI revolution?
It was written by eight researchers at Google.
All eight have now left.
Yes, every single one.
Noam Shazeer departed earlier this year for OpenAI, just two years after Google spent nearly $3 billion to bring him back. Shortly afterward, Nobel Prize winner John Jumper left DeepMind for Anthropic.
Why are they leaving?
Two reasons stand out.
The first is bureaucracy. Layers of approvals often stand between research and product deployment.
The second is compute access. Researchers inside Google see TPUs being sold to cloud customers, including Anthropic, which builds models that directly compete with Gemini.
As analyst Gil Luria put it:
"They are not interested in commercializing AI. They want to be part of history."
In other words, these people are not leaving because Google is failing.
They are leaving because they want laboratories, not balance sheets.
WHAT GOOGLE HAS BUILT
While researchers were leaving, Google was building something few competitors can fully match.
It designs its own chips.
It operates its own global network of data centers.
It sells AI models, databases, security software, and AI agent tools.
Cloud revenue grew 82%, significantly faster than major cloud rivals.
According to Sundar Pichai, 90% of Fortune 100 companies already use Gemini Enterprise.
What does that mean?
Google is monetizing AI through three different channels at the same time.
It sells infrastructure to companies such as OpenAI and Anthropic.
It sells Gemini to enterprises.
And it embeds AI into Search, YouTube, and Workspace for billions of users.
There is another important point many investors overlook.
You do not always need the most advanced model in the world.
Investor Dan Niles summarized it well:
"The models are good enough for 90% of what needs to be done. You don't need a Ferrari. A Ford gets the job done most of the time."
Venture capitalist Tomasz Tunguz has made a similar argument, saying that for most office work, reasonably capable models are already sufficient.
Google understands this.
That is why its Gemini Flash models aim to deliver near frontier-level capabilities while running much faster and more efficiently.
However, there is still a challenge.
Gemini 3.5 Pro has been delayed.
And compute resources remain constrained because every TPU allocated to one project is unavailable somewhere else.
WHO IS PAYING FOR ALL OF THIS?
Last Thursday, Alphabet raised another $25 billion through ten separate bond offerings with maturities ranging from two to forty years.
And this was not the first time.
During the first half of the year alone, the company issued more than $50 billion in new debt and plans to continue doing so on a regular basis.
Why?
To help finance roughly $205 billion in AI infrastructure investments.
And Google is not alone.
Since 2025, Amazon, Alphabet, Nvidia, Meta, Oracle, and SpaceX have collectively raised more than $350 billion through bond markets, with Amazon and Alphabet leading the pack.
So if you are wondering why Google's stock fell despite strong results, the answer is simple.
The issue is not revenue.
The issue is cost.
