On the Red Hat Summit 2026 show floor in Atlanta, I stopped at the Lenovo booth and spent a good chunk of time with Jessie Lacome, Platform Solutions Product Manager at Lenovo. She walked me through the Lenovo ThinkSystem SR675 V3, and the more of the spec sheet she went through, the more it read like a deliberate answer to a very specific question: how do you sell one box to a customer who wants to consolidate virtual machines today and doesn’t want to buy a second box for AI next year?
The Spec Sheet, and Why Each Line Is a Buying Decision
The SR675 V3 is a dual-socket AMD EPYC server with 192 cores per socket — 384 cores total before you even count simultaneous multithreading. For anyone sizing an OpenShift Virtualization cluster, core count is the number that directly sets your VM density per node. More cores per physical host means fewer hosts to patch, fewer hypervisor licenses, and a smaller failure domain to plan HA around. Going wide with EPYC’s core counts rather than clocking a handful of cores higher is the right trade for consolidation workloads, where most VMs are waiting on I/O or memory, not chewing through single-thread performance.
The 4TB of system RAM matters for the same reason. Virtualization density lives or dies on the memory ceiling, not the CPU — you run out of RAM to hand out to guests long before you run out of cores on a modern EPYC part. 4TB gives serious headroom for dense VM consolidation, and it’s also enough to keep large in-memory datasets and model-serving caches resident without going back to the CPU-GPU interconnect for every request.
Then there’s the detail that’s easy to skim past: it’s fully air-cooled. That’s a real buying decision, not a footnote. Liquid cooling in dual-width, high-TDP GPU servers usually means a facility retrofit — cooling distribution units, leak detection, plumbing to the rack — which is a hard sell for a customer who wants to buy a server, rack it, and turn it on. An air-cooled chassis that still supports dual-width GPUs at this power envelope tells you Lenovo engineered the airflow and chassis depth specifically so a mainstream data center doesn’t need a facilities project just to say yes to AI. For a lot of enterprise buyers, that’s the difference between a purchase order this quarter and a capital project next year.
Dual-Width AMD Instinct MI350X: What the GPU Slot Signals
The SR675 V3 supports dual-width AMD Instinct MI350X GPUs, plus rear-mounted networking and extra PCIe slots for expansion. Dual-width is the physical form factor that carries the memory and power budget these accelerators need — it’s the same design constraint NVIDIA’s data center parts impose, and Lenovo building a chassis around it for AMD’s accelerator line is a signal worth reading. AMD Instinct in a mainstream Lenovo SKU, sitting next to Lenovo’s existing NVIDIA-based lines, means enterprise buyers now get a real second source for GPU-accelerated servers — not just a spec-sheet mention, but a shipping, air-cooled, dual-socket box you can configure today. For procurement teams that have spent the last two years worried about GPU supply concentration, that optionality matters as much as the raw FLOPS.
Rear-mounted networking and the extra PCIe slots round out the story: they’re there so the NICs and GPUs aren’t fighting for the same lane budget, and so there’s room left for storage or DPU cards without re-architecting the chassis. That’s what “expansion headroom” concretely looks like in a 2U-class chassis, not just a marketing phrase.
Positioned as an On-Ramp, Not a Ceiling
What stuck with me from Jessie’s pitch is the framing: the SR675 V3, running Red Hat OpenShift on top, is positioned as a solid entry point into virtualization with plenty of headroom to run all the AI you can consume. That’s a deliberate sequencing argument — buy the box to consolidate VMs now, and the same hardware absorbs GPU-accelerated inference or fine-tuning workloads later without a forklift upgrade. For a buyer weighing a dedicated AI server against a general-purpose virtualization host, this is Lenovo’s answer: don’t choose, buy the box that does both.
If you want to configure and price this exact kit, Jessie pointed me to dcsc.lenovo.com, Lenovo’s direct configure-and-buy portal — worth a look if you’re already scoping a refresh for OpenShift Virtualization and want AI capacity built in from day one rather than bolted on later.
Related Reading
- Red Hat Summit 2026 Atlanta: Open Source AI Recap
- Operationalizing OpenShift Virtualization at Red Hat Summit 2026
- OpenShift Virtualization Storage at Red Hat Summit 2026
- GPU Hardware Selection Guide for RHEL AI
- Red Hat AI Factory with NVIDIA: Confidential Compute at Summit 2026
About the Author
I am Luca Berton, AI and Cloud Advisor. I work at the intersection of platform engineering, cloud security, and enterprise AI deployments. Book a consultation.
