NVIDIA’s next-generation Vera Rubin AI servers are already under development by manufacturing partner Foxconn, with mass production scheduled for the second half of 2026. The cutting-edge NVL144 MGX servers will become Foxconn’s primary AI product from late 2026 through 2027, representing a significant leap in AI computing capabilities. This development comes as NVIDIA continues its rapid acceleration of datacenter technology, moving from Hopper to Blackwell and now Vera Rubin within just 30 months.
Table of Contents
NVIDIA Vera Rubin AI Servers in Development
Foxconn has initiated engineering validation for the Vera Rubin NVL144 MGX liquid-cooled racks, positioning itself as NVIDIA’s main development partner with an estimated 60% market share of AI server shipments. The Taiwanese manufacturer is also heavily investing in US-focused facilities, aligning with NVIDIA’s vision of American manufacturing. Foxconn’s Kaohsiung-1 facility in Taiwan is being built as a 40-megawatt datacenter specifically designed for 800-volt DC power architecture to support these next-generation systems.
| Vera Rubin AI Server Details | Information |
|---|---|
| Configuration | NVL144 MGX liquid-cooled racks |
| GPU Count | 144 Rubin GPUs per unit |
| Memory Type | HBM4 |
| Mass Production | Second half of 2026 |
| Power Architecture | 800-volt DC |
| Cooling | 100% liquid-cooled (45°C) |
| Development Partner | Foxconn (60% market share) |
| Networking | ConnectX-9 800GB/s |
Revolutionary Architecture and Design
The Vera Rubin NVL144 features a modular, 100% liquid-cooled design with a central printed circuit board midplane replacing traditional cable-based connections. This innovation enables faster assembly and easier serviceability while supporting NVIDIA ConnectX-9 800GB/s networking and Rubin CPX for large-scale inference workloads. The rack design incorporates energy-efficient 45°C liquid cooling, a new liquid-cooled busbar for higher performance, and 20 times more energy storage to maintain steady power delivery.
NVIDIA is contributing these upgraded rack and compute tray innovations as open standards to the Open Compute Project consortium, enabling over 50 MGX system and component partners to mix and match components in modular fashion for faster scaling.
Strategic Shift to 800-Volt Datacenters
A cornerstone of the Vera Rubin platform is the transition to 800-volt direct current datacenters, replacing legacy 415-volt AC systems. This approach moves power conversion upstream, delivering DC current directly to racks to reduce energy loss and simplify the electrical path from grid to compute node. Companies like Foxconn, CoreWeave, Oracle Cloud Infrastructure, Lambda, and Together AI are already designing datacenters around this 800 VDC standard, which promises increased scalability, better energy efficiency, and reduced material requirements.
The technology, already proven in electric vehicles and solar power systems, will support what NVIDIA calls “gigawatt AI factories”—massive facilities integrating compute, power, and cooling as unified infrastructure.
Market Timeline and Competition
Currently, GB200 Blackwell AI servers drive revenue for NVIDIA and its partners, with Blackwell Ultra GB300 servers entering mass production and shipments continuing through early 2026. Foxconn has already completed small-scale production of GB300 servers, which became its main AI product in the first half of 2026. OpenAI has reportedly placed substantial preorders for Vera Rubin silicon, securing initial production volumes and supporting Foxconn’s capacity investments.
The rapid 30-month cycle from Hopper to Blackwell to Rubin represents the fastest acceleration in datacenter equipment history, forcing hyperscale operators to speed up depreciation schedules. Industry analysts estimate Foxconn could ship up to 40,000 AI server cabinets by 2028. Learn more about NVIDIA’s AI innovations at NVIDIA’s official website and explore additional technology coverage on Technosports.
FAQs
When will NVIDIA Vera Rubin AI servers begin mass production?
Mass production is scheduled for the second half of 2026, with Foxconn leading manufacturing efforts.
What makes Vera Rubin different from current AI servers?
It features 144 Rubin GPUs with HBM4 memory, 100% liquid cooling, 800-volt DC power, and modular MGX architecture for faster deployment.
