IBM
IBM appears in NVIDIA's system-builder lists and designs its own silicon and Power systems, but its x86 NVIDIA server story is thinner than the other OEMs here. Listed for completeness: most buyers meet IBM's GPU hardware through watsonx or storage, not merchant HGX boxes.
- IBM does not build NVIDIA HGX/MGX/NVL72 GPU compute servers and is not an NVIDIA systems builder in the Dell/HPE/Supermicro sense
- Its real NVIDIA hardware certification is storage: Storage Scale System 6000 is NVIDIA-Certified Storage, extended in 2025 to HGX H100/H200/B200 and GB200 NVL72 clusters as the parallel filesystem
- Power servers dropped NVIDIA GPU attachment after Power9 (the Summit/Sierra era); Power10 and Power11 use IBM's own Spyre Accelerator instead
- Current NVIDIA GPU compute exposure is consumption and resale via IBM Cloud (A100, H100, L40S; Blackwell Ultra planned Q2 2026), not IBM-designed hardware
| Model | Platform | Form factor | Cooling |
|---|---|---|---|
| Storage Scale System 6000 Attached storage, not a GPU server: up to 310 GB/s read, 47PB flash per 42U rack | storage attach (HGX, GB200 NVL72) | 2U enclosure, scalable to 42U racks | air |
| IBM Cloud GPU instances IBM resells NVIDIA GPUs in cloud; underlying hardware is not IBM-designed | PCIe / HGX-class cloud | cloud bare metal | TBD |
| Power AC922 (legacy) Legacy Summit/Sierra generation; NVIDIA attachment ended with Power9 | Power9 + V100 NVLink | 2U | air or DLC |
- NVIDIA-Certified Storage: Storage Scale System 6000 (DGX SuperPOD/BasePOD since Aug 2024, extended to GB200 NVL72)
- NVIDIA Cloud Partner certification for Storage Scale System 6000 (Jan 2025)
- Expanded IBM-NVIDIA collaboration announced at GTC 2026 across analytics, document AI, cloud and consulting; no IBM-branded GPU server announced
- TBD for current-generation storage systems
- Do not classify IBM alongside Dell/HPE/Supermicro as an NVIDIA server builder: its NVIDIA hardware role is certified storage attached to third-party GPU clusters
- Power line now uses IBM's own Spyre chip for AI inference
- Listed on this index for completeness and its presence in NVIDIA's builder lists; the entry documents why it does not qualify as a GPU server designer
Is IBM an OEM, ODM or integrator for NVIDIA AI servers?
IBM sits in the "Global OEMs" tier of our index. IBM does not build NVIDIA HGX/MGX/NVL72 GPU compute servers and is not an NVIDIA systems builder in the Dell/HPE/Supermicro sense
Does IBM design its NVIDIA servers in-house?
IBM appears in NVIDIA's system-builder lists and designs its own silicon and Power systems, but its x86 NVIDIA server story is thinner than the other OEMs here. Listed for completeness: most buyers meet IBM's GPU hardware through watsonx or storage, not merchant HGX boxes.
Which NVIDIA platforms does IBM ship?
IBM ships systems on storage attach (HGX, GB200 NVL72), PCIe / HGX-class cloud, Power9 + V100 NVLink. The systems table on this page lists the specific models.
Where does IBM manufacture its AI servers?
TBD for current-generation storage systems