Foxconn picked by Nvidia as exclusive rack-scale supplier for Groq accelerator: report

Taiwanese firm is a key manufacturer in the Vera Rubin value chain, and said to be ahead of schedule with its efforts

Taiwanese electronics manufacturer Foxconn will reportedly supply components for the inference accelerator underpinning Nvidia's upcoming Vera Rubin platform.

According to Taiwan’s Economics Daily News (EDN), Foxconn is the main supplier of the computing tray and cabinet assembly for Nvidia's Groq 3 LPX. Revealed at GTC 2026, the liquid-cooled LPX racks will power Nvidia’s Groq 3 language processing unit (LPU) offering, with some 256 LPUs interconnected through 640 Tb/s of scale-up bandwidth and 128 gigabytes of on-chip static random access memory.

Groq 3 LPU marks the start of Nvidia's collaboration with Groq, stemming from its pseudo-acquisition-style licensing deal with the AI inference chip designer. The offering is designed to power low-latency AI inference workloads as part of Nvidia’s upcoming Vera Rubin platform, with the LPU seemingly replacing the previously mooted Rubin CPX, a graphic processing unit (GPU) acting as a context processing accelerator for token-heavy workloads.

EDN’s report claims the Groq 3 LPU is shipping ahead of schedule in the third quarter, with an initial shipment of about 6,000 racks, and another 10,000 in 2027. In its November earnings, Foxconn Chairman Young Liu said that the company produces more than 1,000 data center cabinets per week, with plans to increase its production capacity to 2,000 cabinets by the end of this year.

Foxconn, which is known locally as Hon Hai Precision Industry, in its most recent earnings report, saw revenues for the first two months of 2026 jump 22% year-on-year, with an accumulated revenue of $42 billion. The firm has been a key server supplier for Nvidia since around 2009, manufacturing Nvidia-designed server nodes and integrating them into full rack-scale systems, including power, cooling, and networking.

It was revealed last year that the pair are working with the Taiwanese government to build an AI supercomputer. The supercomputer will consist of 10,000 Nvidia Blackwell Ultra GPUs, with AI infrastructure supplied by Foxconn subsidiary Big Innovation Company.