Several current and former Google DeepMind employees told reporters that declining morale is contributing to delays in the lab’s larger AI models. Sources spoke on the condition of anonymity citing fear of retaliation.
Why this matters
The issue has drawn attention because Google’s free cash flow has turned negative while the company is investing heavily in AI. Investors and competitors are increasingly questioning Google’s position in the AI race.
State of play
According to Bloomberg, Gemini 3.5 Pro, DeepMind’s most powerful model in development, is months behind schedule. At the same time, the company recently released a set of smaller, more cost-efficient models that received mixed reactions.
Leaders at competing labs were quick to needle the launches; Meta’s Alexander Wang wrote on X, “Gemini who?”, and others have pressed Google’s senior leadership on when more powerful models will be ready.
Financial backdrop and figures
Although Google beat estimates in its most recent earnings, its free cash flow turned negative—largely due to AI investments. The company is on track to spend about $190 billion this year. Cloud revenue grew by 82%, but search revenue came in slightly below Wall Street expectations, raising concerns about the payoff from AI capital expenditures.
Workforce and morale
Sources say the ongoing talent war in AI has hit Google especially hard, with several senior researchers leaving for rival labs. Departures named by sources include Noam Shazeer, a co-lead on Gemini who is now at OpenAI, and John Jumper, the Nobel Prize in Chemistry winner who moved to Anthropic. Multiple employees have publicly stated they resigned over Google’s April deal with the Pentagon that allows military use of its technology; anonymous sources said the agreement frequently comes up during exit interviews.
Some staff linked the morale decline directly to ethical opposition to the Pentagon deal, while others described burnout stemming from a feeling of being one step behind competitors. One employee told Axios bluntly: “We're behind.”
Former DeepMind researcher Alex Turner resigned over Google’s military contracts and noted that Demis Hassabis—who published an AI safety framework last week—does not have the same consistent internal presence with employees that other AI CEOs, like Sam Altman and Dario Amodei, appear to have.
Sources described a “constant battle” within the company for those morally opposed to the Pentagon agreement, contributing to emotional burnout.
Google’s response
Google disputes the notion that morale issues are causing model shortfalls or that mass departures stem from the Pentagon agreement. The company says attrition rates for AI talent in the first half of this year are lower than they were at the same time last year, and that more than 90% of those offered AI roles at Google accept them.
A Google spokesperson told Axios the company is “feeling good about this week’s Flash launches, our roadmap and the incredible demand we’re seeing for our models.”
Bigger picture
Cost is becoming a major concern for enterprises facing AI sticker shock. While the newly released Google models are not at the absolute performance frontier, offering models that deliver more value per dollar addresses a clear market need.
Model providers leapfrogging one another is normal in the AI race; leads often prove fickle and short-lived. Last year Google was in a strong position, then Anthropic’s Mythos shifted the landscape. Now OpenAI appears positioned to introduce new models that could again change the competitive order.
The bottom line: Google can withstand being leapfrogged technologically, but it cannot afford to lose critical talent or the patience of its investors.
