The postponement of Google's Gemini reveals a more significant issue: employee morale.

      Google is falling behind in a race it once dominated, and the issue may lie with its own employees. According to six current and former DeepMind staff members who spoke to Axios, low morale is hampering the lab’s model releases. The immediate concern is Gemini 3.5 Pro, which Bloomberg reports is months behind schedule. Although Google released a series of smaller, more affordable models this week, the reception was mixed. Competitors quickly jumped in, with Meta's Alexandr Wang posting on X, "Gemini who?"

      This situation is particularly painful as financial resources are dwindling. While Google surpassed earnings expectations on Wednesday, its free cash flow went negative due to AI expenditures, with a forecasted spending of $190 billion this year. Despite an 82% growth in cloud revenue, search performance fell short, raising doubts about the return on capital expenditures.

      A talent exodus is also a significant issue. The battle for AI talent has hit Google hard, resulting in the loss of high-profile individuals. Notably, Gemini co-lead Noam Shazeer has moved to OpenAI, and Nobel Prize-winning chemist John Jumper has joined Anthropic. An employee remarked to Axios that the company's models are lagging behind those of competitors.

      A recurring point of contention is an April agreement allowing the Pentagon to use Google's technology. Several employees indicated that they resigned over this, and sources noted it frequently arises in exit interviews. Those opposed described it as a "constant battle" contributing to "emotional burnout." Former DeepMind scientist Alex Turner, who left due to the military collaboration, remarked that CEO Demis Hassabis does not have the same strong internal presence as Sam Altman and Dario Amodei have with their teams.

      In response, Google contests this depiction. The company argues that morale is not hindering its models and that departures over the Pentagon deal are not widespread. It claims turnover among AI staff is lower than it was a year ago, with more than 90% of individuals offered AI roles choosing to accept them. A spokesperson mentioned, "We’re feeling good about this week’s Flash launches, our roadmap, and the incredible demand." Hassabis has also introduced an AI safety framework recently.

      There is validity in both perspectives. Leadership in AI can be unpredictable, and Google has previously been surpassed, notably by Anthropic’s Mythos, and may soon be outpaced by OpenAI again. However, the lower-cost models it recently unveiled do address a significant market demand regarding expenses.

      Ultimately, the situation is clear-cut. While Google can cope with being overtaken on a leaderboard, it cannot afford to lose its top talent or its investors’ patience.