MiTAC Computing has brought an open, full-stack AI infrastructure proposal — spanning servers to management software — to OCP Global Summit 2026, with two hardware debuts: the C2611Z6, a four-node server with AMD EPYC 9006 and liquid cooling for OCP racks, and a new G4520G6 configuration for inference with up to eight NVIDIA GPUs. The subsidiary of Taiwan’s MiTAC Holdings announced it on October 12, 2026, in San Jose, California, where it is also showing a two-phase cooling prototype with ZutaCore and a modular data center developed with Tonomia.
MiTAC Computing at OCP 2026 key takeaways in 30 seconds
- The C2611Z6 packs four nodes with AMD EPYC 9006 into 2U, with liquid cooling for OCP ORv3 racks.
- The G4520G6 supports up to eight NVIDIA GPUs for enterprise inference with Intel Xeon 6.
- A prototype with ZutaCore uses waterless two-phase cooling for chips above 4,000 W.
- The TonoForge modular data center deploys in 12 weeks, according to MiTAC.
- The company is betting on firmware and management built on open standards.
MiTAC’s message is that large AI clusters aren’t solved with compute power alone. The company, which also owns the TYAN server line and Intel’s former Data Center Solutions Group (DSG) server business, and which previously strengthened its alliance with AMD at Computex 2025, is positioning itself as a full-stack provider: compute and storage servers, liquid cooling, open firmware, operating system, AI software, and pod management. The proposal spans everything from large rack-scale AI factories to modular data centers for fast or decentralized deployments.
Two New Servers for OCP and Inference
The C2611Z6, which MiTAC is showing for the first time, is a multi-node server built for OCP ORv3 racks. It packs four compute nodes into 2U with AMD EPYC 9006 series processors, liquid cooling, 16 DDR5-8000 memory slots per node, and OCP 3.0 expansion. The other debut is the G4520G6E3BC-N configuration of the G4520G6, a 4U GPU server with two Intel Xeon 6 processors that supports up to eight NVIDIA RTX PRO 4500, RTX PRO 6000 Blackwell Server Edition, or H200 GPUs, with 12 NVMe E3.S bays for data-heavy inference workloads.
| Model | Form factor | Processor | Highlights |
|---|---|---|---|
| C2611Z6 | 2U, 4 nodes, liquid cooling | AMD EPYC 9006 | 16 DDR5-8000 slots per node, OCP 3.0, ORv3 racks |
| G4520G6 (G4520G6E3BC-N) | 4U, dual-socket | Intel Xeon 6 | Up to 8 NVIDIA GPUs, 12 NVMe E3.S bays |
| G4826Z5 | Liquid-cooled AI server | Not detailed (AMD Instinct MI355X GPUs) | Basis for the TonoForge modular data center and the ZutaCore prototype |
The G4826Z5 ties together much of the rest of the announcement. It uses AMD Instinct MI355X GPUs with the open ROCm software stack and runs on Canonical-certified Ubuntu 26.04, with Kubernetes and the DDN Data Intelligence platform. MiTAC is working with Rafay on GPU cluster management to offer GPU-as-a-service (GPUaaS) resources allocated according to workload.
Two-Phase Cooling and Modular Data Centers
With ZutaCore, MiTAC is presenting a proof of concept that pairs the G4826Z5 with ZutaCore’s HyperCool waterless two-phase liquid cooling technology. A dielectric fluid turns from liquid to gas directly on the chip, pulling heat away at the source. MiTAC says the system supports more than 4,000 W per chip and an average heat flux above 250 W/cm², with an annualized model pointing to a pPUE of up to 1.03 under defined operating conditions. That’s an estimate, not a production measurement, but it illustrates why liquid cooling has become essential for AI.
For decentralized deployments, MiTAC is showing TonoForge, a modular data center developed with Tonomia that integrates liquid-cooled G4826Z5 servers at rack scale. The company says one deployment was completed in 12 weeks, though it doesn’t detail that installation’s size or location.
Open Firmware to Reduce Vendor Lock-In
On the software side, MiTAC uses two in-house developments as part of its open-firmware push: MiOPF, for fast system boot, and MiOBMC, for real-time hardware monitoring and telemetry. The company argues this gives its customers more control over firmware and reduces dependence on a single vendor. Its MiCoreView management software, built on open-source technologies, extends that management from the server to the rack and the cluster. MiTAC is also holding a private meeting on open infrastructure management with Open Platform Firmware, OpenBMC, and Redfish. The announcement includes no pricing or availability dates for the new systems.
Frequently Asked Questions
What is the MiTAC C2611Z6?
It’s a 2U server with four compute nodes, AMD EPYC 9006 processors, and liquid cooling, designed for OCP ORv3 racks.
What GPUs does the MiTAC G4520G6 support?
Up to eight NVIDIA RTX PRO 4500, RTX PRO 6000 Blackwell Server Edition, or H200 GPUs, with Intel Xeon 6 processors.
What is ZutaCore’s two-phase cooling?
A waterless system in which a dielectric fluid evaporates directly on the chip to pull heat away. MiTAC says it supports more than 4,000 W per chip.
What is TonoForge?
A modular data center from MiTAC and Tonomia with liquid-cooled G4826Z5 servers that the company says deploys in 12 weeks.
Sources:
