Inventec
Inventec designs server boards and HGX systems, mostly landing in hyperscaler fleets under other names. One of the quieter Taiwanese ODMs, but the engineering depth is real: its boards carry a meaningful share of the world's AI compute.
- Builds full racks, not just boards: Artemis rack line for GB200/GB300 NVL72, MGX 1U-architecture compliant, performing motherboard assembly and full-rack integration (L10/L11)
- HGX Partner Program member; designed its own switchboard (host-to-GPU GPUDirect RDMA bridging) and retimer board for HGX 8-GPU systems
- Liquid cooling: closed-loop single-phase cold-plate DLC, sidecar and row-based DLC with a self-developed rack-mounted CDU with leak detection, plus immersion R&D; chairman claims industry-leading liquid-cooling patent count
- One of Dell's top three global server assembly partners
| Model | Platform | Form factor | Cooling |
|---|---|---|---|
| P8000IG6 Supports HGX H100/H200/B200 8-GPU configurations | HGX B200 | 8U | air |
| P9000G7 Paired with Intel Xeon 6 | HGX B300 | 10U | air |
| P5800G7 Liquid-cooled HGX B300 platform | HGX B300 | 5U | DLC |
| Artemis / Artemis II Up to 136kW per rack, sidecar or row-based DLC with rack-mounted CDU | GB200 NVL72 / GB300 NVL72 | full rack | DLC |
| Artemis III Early Vera Rubin rack shown at GTC 2026 | Vera Rubin NVL72 | full rack | DLC |
- NVIDIA HGX Partner Program member with early reference-architecture access
- Named GB200 NVL72 rack builder; GB300 NVL72 builder reportedly gaining order share
- Showed an early Vera Rubin NVL72 rack (Artemis III) at GTC 2026
- Partnered with NVIDIA on IGX Thor infrastructure for physical AI
- Taoyuan, Taiwan
- Ciudad Juarez, Mexico: $450M Megasite, 45 production lines, ~35% of global server output
- Katy, Texas: new facility growing from $85M to a planned $251M across three phases
- Blucina (Brno), Czech Republic: new 52,000 sqm facility (Sept 2025)
- Pure ODM: one of Dell's top three server assembly partners, benefiting from xAI's ~$6B order shift from Supermicro to Dell
- Smaller AI-server ODM than Foxconn/Quanta/Wiwynn but gaining share; 2026 capex guided to roughly double
- US and Mexico expansion framed as tariff mitigation and proximity to US datacenter demand
Is Inventec an OEM, ODM or integrator for NVIDIA AI servers?
Inventec sits in the "ODM system designers" tier of our index. Builds full racks, not just boards: Artemis rack line for GB200/GB300 NVL72, MGX 1U-architecture compliant, performing motherboard assembly and full-rack integration (L10/L11)
Does Inventec design its NVIDIA servers in-house?
Inventec designs server boards and HGX systems, mostly landing in hyperscaler fleets under other names. One of the quieter Taiwanese ODMs, but the engineering depth is real: its boards carry a meaningful share of the world's AI compute.
Which NVIDIA platforms does Inventec ship?
Inventec ships systems on HGX B200, HGX B300, GB200 NVL72 / GB300 NVL72, Vera Rubin NVL72. The systems table on this page lists the specific models.
Where does Inventec manufacture its AI servers?
Taoyuan, Taiwan. Ciudad Juarez, Mexico: $450M Megasite, 45 production lines, ~35% of global server output. Katy, Texas: new facility growing from $85M to a planned $251M across three phases. Blucina (Brno), Czech Republic: new 52,000 sqm facility (Sept 2025)