← 返回首页

Google Cloud Next 2026:TPU 8与$750M Agent基金 Google Cloud Next 2026: TPU 8 & $750M Agent Fund

2026-04-27 · 商业动态

核心信息:Core Info: Google Cloud Next 2026在拉斯维加斯举行,发布TPU 8系列(首次分拆训练/推理芯片)、Gemini Enterprise Agent平台、$750M Agent基金。Sundar Pichai宣布"Agentic Era"到来。 Google Cloud Next 2026 held in Las Vegas, released TPU 8 series (first split training/inference chips), Gemini Enterprise Agent platform, $750M Agent fund. Sundar Pichai declares "Agentic Era" arrival.

一、TPU 8系列:首次分拆训练与推理1. TPU 8 Series: First Training/Inference Split

这是Google首次将旗舰加速器分拆为两个专用变体:

This is Google's first time splitting flagship accelerator into two specialized variants:

TPU 8t (训练)

面向训练场景For training scenarios

  • 单集群可达9,600片Up to 9,600 chips per cluster
  • 2 PB共享HBM2 PB shared HBM
  • 近3倍于Ironwood算力~3x Ironwood compute
  • 前沿模型开发从数月缩短到数周Frontier model from months to weeks

TPU 8i (推理)

面向推理场景For inference scenarios

  • 288GB HBM288GB HBM
  • 384MB on-chip SRAM(3倍于前代)384MB SRAM (3x predecessor)
  • 整个模型工作集可在片上运行Full model on-chip
  • 多Agent工作负载超低延迟Ultra-low latency for multi-agent

二、$750M Agent基金2. $750M Agent Fund

Google承诺投入$7.5亿到其12万成员的合作伙伴生态,加速Agentic AI开发。

Google committed $750 million to its 120,000-member partner ecosystem to accelerate Agentic AI development.

三、Gemini Enterprise Agent平台3. Gemini Enterprise Agent Platform

四、Sundar Pichai的宣言4. Sundar Pichai's Declaration

"我们正处于'Agentic Era'的门口。AI不再只是回答问题,而是能够帮助我们完成复杂任务的智能助手。"

— Sundar Pichai, CEO of Google

五、对工程师的启示5. Implications for Engineers

学到了什么?What did we learn?
  1. 芯片战争升级:Chip war escalates:Google首次分拆TPU,专注于训练和推理的不同需求。这预示着AI硬件的定制化趋势。Google splits TPU for first time, focusing on different training/inference needs. Signals AI hardware customization trend.
  2. Agent平台竞争白热化:Agent platform competition intensifies:Google入局Enterprise Agent平台,与OpenAI Workspace Agents、Microsoft Copilot Studio展开正面竞争。Google enters Enterprise Agent platform, competing directly with OpenAI Workspace Agents, Microsoft Copilot Studio.
  3. 多Agent成为标配:Multi-Agent becomes standard:TPU 8i专门为多Agent工作负载优化,说明多Agent协作已是大势所趋。TPU 8i optimized specifically for multi-Agent workloads, signaling multi-Agent collaboration is the trend.

来源:Source: Baus AI