# Agent 业务承载专题日跟踪 · 2026-09-11

> 日期：2026-09-11（Asia/Shanghai）。专题正文：`intel/topics/agent-hosting.md`。原料：`archive/2026-09-11/`。

## 摘要

- **新稿 · Nutanix NAI 2.8 / Agentic（GEC 09-09）：** MCP Gateway GA；Private Inference（LoRA <8B、投机解码宣称至 2.5×——**厂商自称**）；防 rogue AI 的细粒度 IAM；NKP 上 Flow 网络沙箱叙事；dual-native（VM+容器+AI 同控）。
- **续档 · RH AI 3.5 + OpenShell DP + OpenClaw BYOA + Rafay×RH：** 生产 Agent 四问（信任/控制/构建/计量）+ 内核沙箱（标 DP）+ 主权 AI 云参考架构——见 09-10 详摘，本日 archive 再抓。
- **规划：** Agent 承载对比表应并行看 **Nutanix MCP Gateway GA** 与 **RH OpenShell DP**（成熟度不可混写）。

## 原文摘录

### Nutanix NAI 2.8 · MCP / Private Inference

- 归档：`industry/nutanix-agentic-ai-gec-2026-09-09.txt`
- URL：https://gecnewswire.com/nutanix-simplifies-production-agentic-ai-with-new-platform-updates/

> "Enterprise AI should not require customers to rebuild the systems that already run their business," said Thomas Cornely…
> Agent Gateway: This now includes a generally available MCP Gateway…
> Speculative decoding which accelerates LLM inference token generation by up to 2.5x…（**厂商自称**）
> … fine-grained Identity and Access Management (IAM)… restricting access to only authorized roles…

### RH AI 3.5 / OpenShell · 续档

- `archive/2026-09-11/redhat/red-hat-ai-3-5-scaling-and-governing-ai-agents-in-production-0f281ed9.txt`  
  > Moving beyond isolated development sandboxes requires answering 4 questions before scaling: Can we trust it? Can we control it? Can we build on it? Can we measure it?
- OpenShell DP：`archive/2026-09-10/redhat/beyond-container-boundaries-kernel-level-agent-security-in-red-hat-ope-42e5502f.txt`  
  > … ships OpenShell agent sandboxing as a Developer Preview…

## 检索范围

- archive/2026-09-11/industry（Nutanix）、redhat；对照 2026-09-10 agent-hosting 日跟踪