Google’s Gemini delay exposes a deeper problem: employee morale

Google is losing a race it once led, and the problem may be its own people. Six current and former DeepMind employees told Axios that low morale is slowing the lab’s model releases, Axios reported.

The immediate symptom is Gemini 3.5 Pro. Google’s most powerful model is months behind, according to Bloomberg. Google did ship a batch of smaller, cheaper models this week, to mixed reviews. Rivals pounced. “Gemini who?” Meta’s Alexandr Wang posted on X.

The timing stings because the money is flowing out. Google beat earnings on Wednesday, but its free cash flow turned negative, largely on AI spending. It is on track to spend $190bn this year.

Cloud revenue grew 82%, yet search came in light, sharpening the question of whether the capex will pay off.

A talent exodus

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol' founder Boris, and some questionable AI art. It's free, every week, in your inbox. Sign up now!

The deeper issue is who is leaving. The AI talent wars have hit Google hardest, and it has lost big names. Gemini co-lead Noam Shazeer went to OpenAI. Nobel chemistry laureate John Jumper left for Anthropic. “We’re behind,” one employee told Axios, of the company’s models against rivals.

One flashpoint keeps surfacing: an April deal letting the Pentagon use Google’s technology. Several employees say they resigned over it, and sources said it comes up often in exit interviews. For those opposed, one called it a “constant battle” that has led to “emotional burnout.”

Alex Turner, a former DeepMind scientist who quit over the military work, said chief executive Demis Hassabis lacks the internal presence that Sam Altman and Dario Amodei have with their staff.

Google pushes back

Google rejects the picture. It disputes that morale is denting its models, or that people are leaving in droves over the Pentagon deal. Attrition among AI staff, it says, is lower than a year ago, and more than 90% of those offered AI roles accept.

“We’re feeling good about this week’s Flash launches, our roadmap and the incredible demand,” a spokesperson said. Its chief executive Hassabis also published an AI safety framework last week.

There is truth on both sides. Leads in AI are fickle, and Google has been leapfrogged before, by Anthropic’s Mythos, and may be again by OpenAI next. The cheaper models it just shipped answer a real market worry about cost.

But the bottom line is stark.

Google can afford to be overtaken on a leaderboard. It cannot afford to lose its best people, or its investors’ patience.