# 产业日报 2026-09-29 来源范围:`/workspace/intel/archive/2026-09-29/`(industry/ 主;vmware/redhat 等 vendor 仅旁证)。抓取统计:Saved 265 / skipped 29 / errors 40;blocked_hosts:blocksandfiles.com、docs.redhat.com、e.huawei.com、support.huawei.com、www.nextplatform.com、www.reddit.com、www.theregister.com。与 `industry/daily/2026-09-28.md` 对照:**industry 本日无公开战略级 HCI/服务器虚拟化产业新稿**;仅见 **OpenAI Custom GPT→Plugins 迁移轻信号**(VR 文日 09-28;公告 09-11;**非 HCI 战略轴,勿抬**)。**续档主轴仍是 Gartner Server Virtualization Platforms MQ(VR 09-23 媒体转述,本地无 PDF)+ VMware 双 MQ/CC 厂商自称(09-24)**;EPYC 9006 Venice(09-28 中强 NEW)→ **本日续档**;ASRock 8005 / Snapdragon 8 Elite Gen6 / X2 Linux / ASUS Seoul = **续档轻**。**Vendor 旁证 NEW(点名;详情归厂商/专题):**(1)VMware **VVF 9.1 Cost Case**(Memory Tiering / vSAN / ops TCO 自称数字)+ **AI Deployments Part2**(GPU utilization / TCO——厂商性能测试博客);(2)RH **OpenShell secure agent onboarding** + **securing AI agents** 系统周边(OpenShell 仍 **开源/DP 叙事,勿写成 GA**)。Register / Blocks&Files / NextPlatform 空壳、至顶网/奇客/科技行者壳、Reddit host_blocked → **跳过**。数字标「厂商自称」「媒体转述」「续档」。 ## 摘要 今日产业信号可以收成五句话: 1. **industry 轻 NEW · OpenAI Custom GPT 退役→Plugins(非 HCI 轴):** VR 转述 OpenAI 公告(09-11):Custom GPT 退役目标 **2026-12-11**;Enterprise 新建截止约 **10-26**(有批准延期可至 **2027-02-11**);改向 **plugins / skills / MCP**,跨 ChatGPT 与 Codex——**开发者平台迁移信号,勿抬成超融合/虚拟化战略日**。 2. **续档主轴 · Server Virt MQ(VR 09-23)+ VMware 双 MQ/CC 自称(09-24):** 全象限仍以媒体转述为准;**本地无 PDF**;勿把 CC 六 #1 写成已核验分数。EPYC 9006 Venice / ASRock 8005 / 手机 SoC Gen6 / X2 Linux / ASUS Seoul / Blueprint / 工业 AIoT / DSX / Open Secure AI→LF / Rubrik×CrowdStrike / IBM skills = **再抓/续档**。 3. **Vendor 旁证 NEW · 成本/推理轴:** VMware **「The Cost Case Is Clear」VVF 9.1**——NVMe Memory Tiering 自称 offload **20–25%** 访存、server TCO 降最多 **42%**;vSAN 相对外置阵列 storage TCO 降最多 **39%**;Quick Patch **30→<5** 分钟(**80%**);诊断工作量最多 **60%**——**全部 Broadcom internal testing / 厂商博客,非独立审计**。并行 **AI Deployments Part2**:passthrough / vGPU(time-sliced / MIG)巩固 GPU SM 利用率(实验自称 ~**34–38%→49%→79%→94%** 等——**厂商性能测试,Llama3.1 8B 场景**)。 4. **Vendor 旁证 NEW · Agent/OpenShift 轴:** RH **Why Red Hat is building secure agent onboarding**——把 **OpenShell** 定位为开源安全 Agent 运行时,并称属 NVIDIA **Open Agent Safety Platform**;Landlock/seccomp/netns/L7、凭证在边界注入、在 Podman 与 **OpenShift** 上验证——**开源/平台能力叙事,勿写成 OpenShell GA 或 OCP Virt 小版本**。并行 **Securing AI agents requires securing the systems around them**——强调身份/工具/运行时/网络/OS 纵深,OpenShell + BlueField 为独立执行点旁证。 5. **对规划:** 早报主抬仍是 **「MQ 媒体全象限 + 许可经济学/迁移 TCO」**;本日叠加 **VVF 9.1 Cost Case(厂商自称 TCO 数字)** 与 **OpenShell 安全 Agent onboarding(DP/开源)** 旁证;OpenAI 插件迁移与手机 SoC **勿抬**;Venice **续档不重复抬中强**。 ## 原文摘录 ### 1. industry 轻 NEW · OpenAI Custom GPT→Plugins(非 HCI) - 归档:`industry/openai-to-retire-custom-gpts-replace-them-with-plugins-7a013a50.*` - URL:https://virtualizationreview.com/articles/2026/09/28/openai-to-retire-custom-gpts-replace-them-with-plugins.aspx - 来源:Virtualization Review;`fetched_at` 2026-09-29T00:42:02Z - 口径:**媒体转述 OpenAI 公告(09-11)**;开发者/企业 ChatGPT 能力迁移;**非 HCI/服务器虚拟化战略轴**。 > … retire **Custom GPTs** … plugin architecture spanning **ChatGPT** and … **Codex**… > … announced … **Sept. 11** … Custom GPT retirement for **Dec. 11, 2026**… > … Enterprise … **Oct. 26** for ending creation of new Custom GPTs… approved deferrals … until **Feb. 11, 2027**… > … plugin … one or more **skills** … **Model Context Protocol (MCP)** server… > … 「A migrated plugin may respond differently」… ### 2. 续档主轴 · Server Virt MQ(VR)+ VMware 双 MQ/CC 自称 #### 2a. Gartner Server Virt MQ · VR 媒体转述(续档) - 归档:`industry/gartner-broadcom-s-vmware-deal-has-reinvigorated-server-virtualization-d8a4a089.*` - URL:https://virtualizationreview.com/articles/2026/09/23/gartner-broadcoms-vmware-deal-has-reinvigorated-server-virtualization.aspx - 口径:**续档**;Leaders:Nutanix(Execute 最高)、Broadcom/VMware(Vision 最右)、Microsoft、Red Hat;Challengers:HPE、Canonical、Oracle;Visionaries:SUSE、Platform9、H3C、IDT、Sangfor;Niche:Proxmox、Vates、VergeIO、Citrix、Scale。**本地无 PDF。** #### 2b. VMware/Broadcom · 双 MQ Leader + CC 六用例 #1(续档 · 厂商 PR) - 归档:`vmware/broadcom-vmware-recognized-as-a-leader-in-two-2026-gartner-magic-quadr-8b97b53f.*` - URL:https://blogs.vmware.com/cloud-foundation/2026/09/24/broadcom-vmware-recognized-as-a-leader-in-two-2026-gartner-magic-quadrant-reports/ - 口径:**续档**;DHI + Server Virt Leaders 自称 + CC 六用例全 #1 自称——**本地无 PDF、未独立核验 CC**。 ### 3. Vendor 旁证 NEW · VMware VVF 9.1 Cost Case + AI Deployments Part2 #### 3a. VVF 9.1 Cost Case(成本轴 · 厂商自称) - 归档:`vmware/the-cost-case-is-clear-infrastructure-economics-with-vmware-vsphere-fo-0975a8ca.*` - URL:https://blogs.vmware.com/cloud-foundation/2026/09/28/the-cost-case-is-clear-infrastructure-economics-with-vmware-vsphere-foundation-9-1/ - 来源:VMware Cloud Foundation Blog;`fetched_at` 2026-09-29T00:34:57Z - 口径:**厂商博客 NEW**;TCO 数字标 **Broadcom internal testing(May 2026)** / Forrester TEI(commissioned);DRAM **80–170%** YoY 引 SHI Resource HUB 2026——**均非本站核验**。详情归 **cost / storage** 专题。 > … **vSphere Foundation 9.1**… measurable **TCO** reduction… > … DRAM … surged **80 to 170%** year over year in **2026**… [1] > … **NVMe Memory Tiering**… offloads **20 to 25%** of memory accesses… reduces server TCO by up to **42%** [2]… > … database … up to **15%** better performance compared to … **9.0**… > … vSAN global Deduplication … up to **39%** lower storage TCO … external arrays [2]… > … **250 GiB** of vSAN capacity per core… > … Quick Patching … **30** minutes to under **5** minutes… **80%**… > … diagnostic analysis … up to **60%**… > … storage related unplanned downtime … up to **32%**… Forrester TEI … VCF **9** [3]… > Disclaimer: … Broadcom internal engineering estimates… #### 3b. AI Deployments Part2 · GPU utilization(推理/云原生旁证) - 归档:`vmware/optimizing-ai-deployments-with-vmware-cloud-foundation-part-2-8d0d611d.*` - URL:https://blogs.vmware.com/cloud-foundation/2026/09/28/optimizing-ai-deployments-with-vmware-cloud-foundation-part-2/ - 来源:VMware Cloud Foundation Blog;`fetched_at` 2026-09-29T00:34:24Z - 口径:**厂商性能测试系列 Part2 NEW**;Llama3.1 8B / RTX 6000 Pro 实验配置——**Broadcom Performance 自称,非第三方审计**。详情归 **inference / cloud-native / cost**。 > … Part 2 … Maximizing **GPU Utilization** for Best **Total Cost of Ownership**… > … **DirectPath I/O** (passthrough) and NVIDIA **vGPU**… time-sliced … **MIG**… > … average utilization often … at or below **50%** [1]… > … Config 1 … ~**34-38%** GPU SM utilization… Config 2 … **49%**… Config 3 MIG … **79%**… Offline/MLperf batching … **94%**… > … throughput … **2328** → **9314** → **18621** → **23365** tokens/second…(文中实验序列) ### 4. Vendor 旁证 NEW · RH OpenShell onboarding + securing AI agents #### 4a. Secure agent onboarding / OpenShell(Agent · OpenShift AI 线) - 归档:`redhat/why-red-hat-is-building-secure-agent-onboarding-e125232b.*` - URL:https://www.redhat.com/en/blog/why-red-hat-is-building-secure-agent-onboarding - 来源:Red Hat Blog;`fetched_at` 2026-09-29T00:34:26Z - 口径:**厂商博客 NEW**;OpenShell = **开源项目 + 安全 Agent 运行时**,并称属 NVIDIA **Open Agent Safety Platform**;计划纳入 **Red Hat AI** 原生能力——**勿写成 GA**;历史站立卷宗仍标 OpenShift AI **Developer Preview** 边界。详情归 **openshift / agent-hosting / security**。 > … encoding into **Red Hat AI** with **OpenShell** — an **open source** project and a secure agent runtime also part of the NVIDIA **Open Agent Safety Platform**… > … **Landlock**, **seccomp**, user and network **namespace** isolation, and **L7** inspection… > … Credentials live **outside** the agent's workload… injected only at the network boundary… > … validated … on both **Podman** and **Red Hat OpenShift**… > … integrate OpenShell into **Red Hat AI** as a native platform capability… #### 4b. Securing AI agents · 系统周边(安全/Agent) - 归档:`redhat/securing-ai-agents-requires-securing-the-systems-around-them-79bd2c93.*` - URL:https://www.redhat.com/en/blog/securing-ai-agents-requires-securing-systems-around-them - 口径:**厂商博客 NEW**;纵深防御(身份/工具/运行时/网络/OS);**OpenShift AI / OpenShift / RHEL** 分层;OpenShell + **BlueField** 为独立于 agent 工作负载的执行点——**架构叙事,非 ACS 发行说明**。 > … Securing an AI agent … means securing the **system** in which the agent acts… > … **OpenShift AI**… **OpenShift**… **RHEL**… **Red Hat AI Inference**… > … Agent runtimes such as **OpenShell**… **NVIDIA BlueField** DPUs… > … **Lightwell**… Open Secure AI Alliance… ### 5. 产业/硬件续档(txt 再抓,非今日 NEW 主轴) | 条目 | 归档 | 相对 09-28 | |---|---|---| | Server Virt MQ(VR) | `industry/gartner-broadcom-s-vmware-deal-has-reinvigorated-server-virtualization-d8a4a089.*` | **续档主轴** | | VMware 双 MQ/CC PR | `vmware/broadcom-vmware-recognized-as-a-leader-in-two-2026-gartner-magic-quadr-8b97b53f.*` | **续档** | | EPYC 9006 Venice | `industry/amd-takes-the-lid-off-of-next-gen-epyc-9006-venice-as-zen-6-comes-to-s-35db7b69.*` | **续档**(昨日中强 NEW) | | ASRock EPYC 8005 | `industry/asrock-rack-soranod8-2l2t-review-a-new-amd-epyc-8005-sorano-motherboar-a6ba9697.*` | **续档轻** | | Snapdragon 8 Elite Gen6 | `industry/qualcomm-unveils-snapdragon-8-elite-gen-6-and-elite-extreme-gen-6-next-406621f3.*` | **续档轻** | | Snapdragon X2 Linux | `industry/qualcomm-announces-snapdragon-x2-debian-and-ubuntu-support-in-2026-964fe909.*` | **续档轻** | | ASUS AI Tech Seoul | `industry/asus-ai-tech-2026-seoul-south-korea-917bff97.*` | **续档轻** | | Blueprint / 工业 AIoT / DSX / Open Secure AI→LF / Rubrik×CS / IBM skills | 同前序 industry 路径 | **续档** | | HCX Part3 / Flow disconnected / memtier / RHEL STIG / Friday Five | 同前序 vendor 路径 | **续档**(昨日 NEW) | ### 6. 空壳 / blocked / 跳过 | 条目 | 说明 | |---|---| | The Register | `host_blocked` / 空壳 → **不摘** | | NextPlatform / Blocks&Files | blocked / 空壳 → **不摘** | | 至顶网 / 奇客 / 科技行者频道壳 | **跳过** | | Reddit | host_blocked | | VR「AI speeds science」等调研壳 | 非 HCI 战略主轴,**本日不抬** | ### 7. 对规划一句 **主抬仍是 MQ/许可/迁移 TCO;** 本日叠加 **VVF 9.1 Cost Case(厂商自称 Memory Tiering/vSAN/ops TCO 数字)** 与 **RH OpenShell 安全 Agent onboarding(开源/DP,纳入 Red Hat AI 叙事)**;OpenAI 插件迁移 **勿抬 HCI**;EPYC 9006 **续档**;AI Deployments Part2 GPU 利用率数字 **仅厂商实验旁证**,勿写成合同 SLA。