Gemini 3.5 Pro Delay Exposes Morale Crisis at Google DeepMind

The launch of Gemini 3.5 Pro, the most powerful model Google DeepMind is building, is running months behind its original schedule. According to conversations Axios held with six current and former lab employees, the delay isn't just a technical hiccup but a symptom of growing discontent inside the team, fueled by the departure of key researchers and a controversial deal with the Pentagon.

Six voices from DeepMind

Axios spoke with six people currently or previously affiliated with Google DeepMind, who described the atmosphere inside the lab as increasingly tense. The sources say the drop in morale is directly slowing work on the company's most important models, including Gemini 3.5 Pro, seen as a key piece of Google's response to the widening lead held by OpenAI and Anthropic.

We're behind - a Google DeepMind employee, quoted by Axios

The remark points to Google's perceived competitive standing against its rivals. The newer, smaller and cheaper models the company released this week got a mixed reception, and according to the sources cited, none of Google's current models rank in the top ten of the LLM leaderboard maintained by OpenRouter.

Pentagon deal as flashpoint

Many employees point to Google's April deal with the Pentagon, which gives the military access to the company's technology, as a source of growing division within the team. The topic comes up regularly in exit interviews with departing staff, and one source described the situation as a constant source of tension leading to emotional burnout.

constant battle, emotional burnout - a former Google DeepMind employee on the effects of the Pentagon deal, quoted by The Next Web

Among those who left over objections to working with the military, Axios names researcher Alex Turner. That's not the only high-profile departure from the lab in recent months, though - Gemini architecture co-creator Noam Shazeer moved to OpenAI, and Nobel Prize-winning chemist John Jumper joined Anthropic.

Money and investor pressure

The financial backdrop raises the stakes further. Even though Alphabet beat profit forecasts last quarter, the company's free cash flow turned negative for the first time, and it plans to spend $190 billion on AI infrastructure this year. Cloud revenue grew 82 percent, but the search segment's results came in slightly below analyst expectations.

In such a tense environment, any delay to a flagship model becomes fodder for public mockery from competitors. Meta's Alexandr Wang commented on the situation with a short post on X: 'Gemini who?', referencing the no-show of the long-promised model.

Google's response

The company rejects the idea that staffing problems have had a real effect on its schedule. According to Google, attrition among AI team employees in the first half of 2026 is lower than a year earlier, and more than 90 percent of candidates accept the AI job offers extended to them.

We're feeling good about this week's Flash launches, our roadmap and the incredible demand - Google statement, quoted by The Next Web

What it means for the market

For Polish companies and developers using Google's models, this has practical implications: the delay of Gemini 3.5 Pro means OpenAI's and Anthropic's technical edge on the most demanding tasks could last longer than expected back in spring. It fits a broader pattern in which Google has been losing employees to rivals for months while facing investor questions about the wisdom of massive AI infrastructure spending amid negative cash flow.

The episode also shows that tensions over the ethics of using AI in military contracts, previously visible mainly in public statements from departing researchers, are now directly affecting the pace of product development at the world's leading AI research labs. Sources: Google's Gemini delay exposes a deeper problem: employee frustration (axios.com), Google's Gemini delay exposes a deeper problem: morale (thenextweb.com), Google's Gemini delay exposes a deeper problem (finance.yahoo.com)