---
格式版本: 2
标题: "The Next Evolution of AI Infrastructure: Inside the Architecture Powering the AI Factory Era"
原文链接: "https://news.lenovo.com/inside-the-architecture-powering-the-ai-factory-era/"
发布日期: "2026-07-23"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:scrape:provider_published_at"
发布时间证据: "provider publishedAt: 2026-07-23"
发布时间校准原因: "规则确认唯一严格发布时间，来源 scrape:provider_published_at"
发布时间校准置信度: "high"
发布时间候选数量: 15
发布时间严格候选数量: 3
发布时间原页读取状态: "原页面已读取"
发布时间未找到原因: ""
发布时间校准时间: "2026-07-26T00:32:09+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-07-26T00:26:20+08:00"
入库时间: "2026-07-25T16:32:09.227Z"
来源平台: "固定入口"
搜索渠道: "fixed_url"
搜索词: "https://news.lenovo.com/"
匹配关键词:
  - "GPU"
  - "Scale-up"
  - "Liquid Cooling"
  - "deployment"
  - "performance"
相关厂家:
  - "Lenovo"
  - "AMD"
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "是"
AI打分: 94
AI分档: "高置信优质"
AI质检状态: "通过"
AI打分理由: "直接讨论AI工厂时代机柜级架构，Lenovo官方发布AMD Helios方案，含72GPU、31TB HBM4、2.9EFLOPS性能，明确2026Q4可用，技术细节与商业信号均强。"
AI质检模型: "deepseek-v4-flash"
AI质检时间: "2026-07-27T11:18:23+08:00"
AI主题相关性: 20
AI来源权威性: 15
AI新颖性: 18
AI技术细节: 16
AI商业部署信号: 15
AI完整性: 10
采集批次: "2026年7月25日22点32分16秒"
采集批次ID: "20260725-223216-273"
去重键: "https://news.lenovo.com/inside-the-architecture-powering-the-ai-factory-era"
---

AI factories aren’t the future anymore; they’re becoming the new blueprint for hyperscale AI. As enterprises and cloud providers move beyond deploying individual AI servers toward building AI factories capable of supporting massive inference and training workloads, infrastructure must evolve from standalone systems to integrated rack-scale designs.

That shift took center stage at Lenovo Tech World ’26, where AMD Chair and CEO Dr. Lisa Su announced Lenovo as one of the earliest OEM adopters of the AMD Helios™ rack-scale solution. Lenovo is applying decades of infrastructure engineering expertise to help customers build the next generation of AI factories – from scale-up and scale-out cluster design to the deployment and lifecycle services that keep AI environments running at scale.

Designed for the next era of generative and agentic AI, AMD Helios™ brings together next-generation AMD Instinct™ GPUs, AMD EPYC™ CPUs, AMD Pensando™ networking technologies and the AMD ROCm™ software ecosystem into an open, rack-scale architecture built on Open Compute Project (OCP) and Open Rack Wide (ORW) specifications. Engineered for hyperscale inference, training and fine-tuning, the solution is built to scale from rack-level systems into datacenter-scale clusters for distributed inference and the largest foundation-model training runs.

To support AI at this scale, the platform combines **72 AMD Instinct™ MI455X GPUs**, next-generation AMD EPYC™ processors, **31TB of HBM4 memory**, and AMD Pensando™ networking technologies in a single rack. The result is **up to 2.9 exaFLOPS of FP4 AI inference performance** and **1.4 exaFLOPS of FP8 AI training performance**, giving hyperscalers and NeoCloud providers the performance needed to train, fine-tune, and serve increasingly sophisticated AI models.

“AI infrastructure is rapidly evolving from standalone servers to fully integrated rack-scale clusters,” said Conor Malone, Vice President and General Manager of Cloud Service Providers, Lenovo. “Working with AMD, we’re bringing together world-class cluster design with Lenovo’s expertise in deploying at scale, helping customers leverage this solution to build AI environments that are open, scalable, and ready for what’s next.”

**More Than Hardware: A Faster Path to Deployment**

Preparing organizations for the new era of AI isn’t simply about deploying new hardware. It requires expertise in cluster design, liquid cooling, workload optimization, and lifecycle management. Leveraging its proven experience in deploying large-scale AI infrastructure, Lenovo helps customers accelerate deployment, optimize performance, reduce time to first token, and move AI environments into production faster.

Lenovo’s engineering and services experts work alongside customers to assess workload requirements, right-size deployments, optimize GPU configuration, and tune performance for the workload at hand. From there, ongoing full-lifecycle services capabilities reduce stand-up time and support long-term differentiation with greater speed and efficiency, helping organizations maximize GPU utilization and reduce deployment risk.

The AI factory may be changing how infrastructure is built, but Lenovo’s goal remains the same: help organizations deploy AI where it delivers the greatest value. Through its Hybrid AI vision, Lenovo is building the open, scalable infrastructure that enables customers to run AI across edge, enterprise, cloud, and hyperscale environments—meeting them where they are today while helping them prepare for what’s next. Lenovo solutions based on the AMD Helios rack-scale solution are expected to be available in 4Q 2026.