Foxconn is developing Nvidia's next-generation Vera Rubin server architecture and securing custom chip orders from cloud providers, reinforcing its market-leading position as artificial intelligence infrastructure spending continues despite mixed industry sentiment.
The Taiwanese manufacturer, known as Foxconn, has locked in multiple AI server contracts using cloud service providers' proprietary application-specific integrated circuits starting in 2026, while ramping production of Nvidia's current GB200 series and preparing for GB300 volume increases after the fourth quarter.
Oracle and Alphabet are among the major cloud operators maintaining capex growth for data center expansion, extending the revenue cycle for hardware suppliers even as technology commoditization intensifies competition among manufacturers.
Nvidia's Vera Rubin platform, scheduled for a second-half 2026 launch, will feature an Arm-based CPU with Rubin GPU delivering 50 petaflops of inference performance and 288GB of high-bandwidth memory. The architecture represents a significant advance from current Blackwell designs, targeting large-scale AI data center deployments.
Foxconn's NVL144 system design will use the existing Oberon chassis while introducing modular GPU configurations and enhanced power delivery through 54-volt systems. Cooling will transition entirely to fanless liquid designs to handle increased computing loads, though final specifications depend on testing validation.
The company currently supplies custom semiconductor servers to AWS and Google Cloud, with large-volume ASIC orders expected to surge from 2026. Industry sources note Foxconn maintains a substantial resource investment advantage over competitors, expressing early interest in the platform.
The developments underscore how sustained cloud infrastructure spending is offsetting supply chain margin pressure from maturing AI server technology and expanding manufacturing competition.
Article translated by Vyra Wu and edited by Jack Wu
