AI News AI资讯 15d ago Updated 12d ago 更新于 12天前 66

Google dismantles Deepmind and bets on a fresh start as Hassabis heads for the exit 谷歌解散DeepMind并押注全新起点,创始人哈西斯或将离任

Google DeepMind is losing its independence and being absorbed as a Google subsidiary, with founder Demis Hassabis potentially departing in the coming months Day-to-day operations are shifting to Koray Kavukcuoglu based in the U.S., while co-founder Sergey Brin is expected to take on greater control over AI development Gemini-related AI work is moving to the Bay Area, signaling a centralization of Google's frontier model efforts away from DeepMind's London base Experts are divided on whether this Google DeepMind正在失去独立性,创始人Demis Hassabis可能即将离开,日常运营由Koray Kavukcuoglu从美国接管 AI开发重心从伦敦转移至美国湾区,Google联合创始人Sergey Brin将重新掌握更大控制权 两种解读并存:SemiAnalysis认为这是Google在AI模型竞赛中落后的信号,Tim O'Reilly则认为Google战略性转向基础设施而非模型领导地位 Gemini 3.1 Pro仍处于预览模式,3.5 Pro项目被搁置,暴露Google在前沿AI开发上的困境 Google Cloud预计2027年AI基础设施收入730亿美元、TPU销

75
Hot 热度
65
Quality 质量
70
Impact 影响力

Analysis 深度分析

TL;DR

  • Google DeepMind is losing its independence and being absorbed as a Google subsidiary, with founder Demis Hassabis potentially departing in the coming months
  • Day-to-day operations are shifting to Koray Kavukcuoglu based in the U.S., while co-founder Sergey Brin is expected to take on greater control over AI development
  • Gemini-related AI work is moving to the Bay Area, signaling a centralization of Google's frontier model efforts away from DeepMind's London base
  • Experts are divided on whether this reflects a strategic pivot toward profitable cloud/infrastructure or an admission that Google is falling behind in the frontier AI model race
  • Gemini 3.1 Pro remains in preview and Gemini 3.5 Pro has been shelved, suggesting serious developmental challenges in Google's coding and frontier model capabilities

Why It Matters

This restructuring represents one of the most significant organizational shifts in the AI industry, directly affecting the trajectory of Google's frontier model development and its competitive position against OpenAI, Anthropic, and other leaders. For AI practitioners and researchers, it signals potential changes in resource allocation, research priorities, and the pace of Google's model releases. The move also raises broader questions about whether tech giants should prioritize infrastructure dominance or frontier model leadership in an increasingly competitive landscape.

Technical Details

  • Koray Kavukcuoglu is taking over operational leadership from the U.S. without holding the CEO title, indicating DeepMind no longer has independent executive leadership
  • All Gemini-related AI development is being relocated to the Bay Area, consolidating efforts that were previously split between Google Brain and DeepMind teams
  • Gemini 3.1 Pro remains in preview mode with no public release date, and Gemini 3.5 Pro — announced in May for a June release — has been effectively shelved
  • Google's latest Flash models are optimized for efficiency rather than raw frontier performance, aligning with a strategy focused on practical utility over benchmark dominance
  • Gemini's annualized revenue stood at $12 billion in Q2 2026, while Google Cloud is projected to generate over $73 billion in AI infrastructure revenue and an additional $120 billion in TPU sales by end of 2027

Industry Insight

  • Google's restructuring suggests that even well-resourced organizations can fall behind in frontier AI when bureaucratic caution and compute allocation constraints slow deployment — a cautionary tale for any lab prioritizing process over speed
  • The pivot toward infrastructure revenue (cloud, TPUs) over model leadership may be a deliberate long-term strategy, but it risks ceding the high-value application layer to competitors who own the model stack
  • AI professionals should watch for accelerated releases or unexpected model announcements from Google in 2026-2027, as the current leadership transition may trigger a more aggressive push to reclaim frontier competitiveness

TL;DR

  • Google DeepMind正在失去独立性,创始人Demis Hassabis可能即将离开,日常运营由Koray Kavukcuoglu从美国接管
  • AI开发重心从伦敦转移至美国湾区,Google联合创始人Sergey Brin将重新掌握更大控制权
  • 两种解读并存:SemiAnalysis认为这是Google在AI模型竞赛中落后的信号,Tim O'Reilly则认为Google战略性转向基础设施而非模型领导地位
  • Gemini 3.1 Pro仍处于预览模式,3.5 Pro项目被搁置,暴露Google在前沿AI开发上的困境
  • Google Cloud预计2027年AI基础设施收入730亿美元、TPU销售1200亿美元,而Gemini年化收入仅120亿美元

为什么值得看

这篇文章揭示了Google在AI战略上的重大转折,反映了科技巨头在模型竞赛与基础设施商业变现之间的深层张力。对AI从业者而言,理解Google的组织调整有助于把握行业竞争格局的演变方向。

技术解析

  • Google DeepMind已取消CEO职位,Koray Kavukcuoglu从美国接管日常运营但不担任CEO头衔,标志着DeepMind作为独立实体的终结
  • Gemini系列模型开发受阻:3.1 Pro仍处预览阶段,3.5 Pro(原计划6月发布)已被搁置,尤其在编程性能上未能追上竞争对手
  • Google Flash系列模型优化方向转向效率而非前沿性能,符合Tim O'Reilly提出的"基础设施优先"战略解读
  • 财务数据对比:Gemini年化收入120亿美元 vs Google Cloud AI基础设施收入730亿美元+TPU销售1200亿美元(2027年预期)

行业启示

  • 科技巨头在AI竞赛中面临"模型领先"与"基础设施变现"的战略抉择,Google的重组反映了资源分配的现实压力
  • 组织结构调整往往暴露技术路线的失败,DeepMind的困境提示前沿AI研发需要更敏捷的决策机制和更激进的算力投入
  • 行业可能见证"基础设施赢家通吃"模式的强化,而非单纯的模型能力竞赛,这对创业公司和研究机构的定位策略具有参考价值

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Research 科学研究 Policy 政策 Gemini Gemini LLM 大模型