China domestic (export-restricted)Vendor profile

xFusion

Verified2026-07-15
Systems listed3

The Huawei server business, spun out in 2021 and now China's #2 server vendor by shipments. Its rack-scale liquid-cooling engineering (FusionPoD) is genuinely impressive, but the lineup centers on Huawei Ascend accelerators, not NVIDIA.

What they design in-house01
  • Inherited Huawei's x86/GPU server engineering team and IP when spun out in September 2021
  • Building a ~$306M proprietary Zhengzhou manufacturing hub, bringing liquid-cooled FusionPoD and 256-GPU Super Node rack assembly in-house
  • FusionPoD for AI: proprietary integrated-cabinet design, 100% native liquid cooling, cable-free blind-mating architecture, pPUE 1.06
  • China's #2 server vendor by shipments
NVIDIA systems02
ModelPlatformForm factorCooling
FusionPoD for AI (700 series)
Up to 64 accelerators per rack, pPUE 1.06, 105kW Titanium PSUs
domestic accelerator47U integrated rackDLC
FusionServer 2288H V7
Direct descendant of Huawei's FusionServer line
PCIe2Uair
256-GPU Super Node
Rack-scale AI training pod moving to in-house Zhengzhou manufacturing
domestic acceleratorrack-scale podDLC
NVIDIA relationship03
  • Successor to Huawei's server unit, which shipped NVIDIA GPUs pre-sanctions; current lineup centers on Huawei Ascend accelerators and Kunpeng CPUs, not NVIDIA
  • Not found on the US Entity List (separate legal entity since 2021), but Huawei lineage warrants independent BIS verification before any transaction
Manufacturing
  • Zhengzhou, Henan (new proprietary hub)
  • Shenzhen (engineering office)
Field notes
  • Spun out of Huawei in 2021 to insulate the server business from sanctions exposure
  • IPO on Shenzhen ChiNext in progress, ~RMB 8B raise reported
  • Structurally different from Inspur/H3C/Sugon: domestic-accelerator-first strategy rather than NVIDIA integration
  • Relevant to Western buyers as a rack-scale liquid-cooling design reference, not an NVIDIA hardware source
Questions we get on this vendor

Is xFusion an OEM, ODM or integrator for NVIDIA AI servers?

xFusion sits in the "China domestic (export-restricted)" tier of our index. Inherited Huawei's x86/GPU server engineering team and IP when spun out in September 2021

Does xFusion design its NVIDIA servers in-house?

The Huawei server business, spun out in 2021 and now China's #2 server vendor by shipments. Its rack-scale liquid-cooling engineering (FusionPoD) is genuinely impressive, but the lineup centers on Huawei Ascend accelerators, not NVIDIA.

Which NVIDIA platforms does xFusion ship?

xFusion ships systems on domestic accelerator, PCIe. The systems table on this page lists the specific models.

Where does xFusion manufacture its AI servers?

Zhengzhou, Henan (new proprietary hub). Shenzhen (engineering office)