Foxconn / Ingrasys
Hon Hai's server arm (Ingrasys, Foxconn Industrial Internet) is the volume leader in AI racks: it designs and builds GB200/GB300 NVL72 systems at a scale no one else matches, including for NVIDIA itself. Increasingly willing to sell rack-scale systems direct to large buyers.
- Ingrasys spans component-level design (chassis, boards) through L10 system integration to L12 rack-and-cluster integration, plus liquid cooling
- Secures over 50% of NVIDIA's HGX GPU baseboard orders per TrendForce
- Builds nearly 1,000 AI server racks per week, with ~40% of each rack's components made in-house, including 60-layer PCBs for GPU substrates (chairman statement)
- In-house liquid cooling: rack-level liquid-to-air sidecars and CDUs rated to 1,300kW, with Omniverse digital-twin simulation
- Builds GB200/GB300 NVL72 racks both to NVIDIA reference and custom for hyperscalers; described as Oracle's largest AI-server supplier (Stargate)
| Model | Platform | Form factor | Cooling |
|---|---|---|---|
| Ingrasys GB200 NVL72 rack ~120kW per rack, 5th-gen NVLink | GB200 NVL72 | full rack | DLC |
| Ingrasys GB300 NVL72 rack Developed in collaboration with NVIDIA | GB300 NVL72 | full rack | DLC |
| Ingrasys HGX/MGX systems NVIDIA reference-architecture platforms | HGX B200 / HGX B300 / MGX | server and rack | air or DLC |
| Ingrasys Vera Rubin NVL72 Roadmap platform, not yet shipping | Vera Rubin NVL72 | full rack | DLC |
- Formal NVIDIA partnership since October 2023 to build AI factories on HGX/OVX reference designs
- Qualified NVIDIA pilot-build partner spanning GB200 NVL72 through GB300 NVL72; Vera Rubin NVL72 listed as upcoming
- Joint Hon Hai/NVIDIA Kaohsiung Super Computing Center: 64 racks of GB200 NVL72 (4,608 Blackwell GPUs)
- November 2025: OpenAI and Foxconn announced direct collaboration to co-design Stargate server racks, cooling, cabling and power
- Houston, Texas: ~1M sq ft Ingrasys site ($450M total investment)
- Mount Pleasant, Wisconsin: $569M additional investment approved Nov 2025
- Guadalajara, Mexico: ~$900M plant described as the world's largest GB200-class AI server assembly site, supplying Stargate
- Kaohsiung, Taiwan; Bac Giang, Vietnam (FII targeting ~$20B Vietnam revenue in 2026); Lordstown, Ohio
- Named customers: OpenAI/Stargate (direct co-design), Oracle (largest AI-server supplier), Amazon, Google via Vietnam expansion reporting
- Pure ODM/contract model: no direct-to-enterprise retail channel, but increasingly sells rack-scale direct to large buyers
- Largest and most aggressive US manufacturing buildout among the Taiwanese ODMs
Is Foxconn / Ingrasys an OEM, ODM or integrator for NVIDIA AI servers?
Foxconn / Ingrasys sits in the "ODM system designers" tier of our index. Ingrasys spans component-level design (chassis, boards) through L10 system integration to L12 rack-and-cluster integration, plus liquid cooling
Does Foxconn / Ingrasys design its NVIDIA servers in-house?
Hon Hai's server arm (Ingrasys, Foxconn Industrial Internet) is the volume leader in AI racks: it designs and builds GB200/GB300 NVL72 systems at a scale no one else matches, including for NVIDIA itself. Increasingly willing to sell rack-scale systems direct to large buyers.
Which NVIDIA platforms does Foxconn / Ingrasys ship?
Foxconn / Ingrasys ships systems on GB200 NVL72, GB300 NVL72, HGX B200 / HGX B300 / MGX, Vera Rubin NVL72. The systems table on this page lists the specific models.
Where does Foxconn / Ingrasys manufacture its AI servers?
Houston, Texas: ~1M sq ft Ingrasys site ($450M total investment). Mount Pleasant, Wisconsin: $569M additional investment approved Nov 2025. Guadalajara, Mexico: ~$900M plant described as the world's largest GB200-class AI server assembly site, supplying Stargate. Kaohsiung, Taiwan; Bac Giang, Vietnam (FII targeting ~$20B Vietnam revenue in 2026); Lordstown, Ohio