NVIDIA announced the Groq 3 LPX at GTC 2026 as its latest AI inference accelerator. The company claims the new chip delivers up to 35 times better AI inference performance than previous generations. Supply chain reports indicate Foxconn will serve as the exclusive supplier for Groq 3 LPX computing trays. This partnership positions Foxconn as a primary manufacturing partner for the LPX accelerator line alongside NVIDIA.
The Groq 3 LPX architecture supports racks containing 256 chips. Each rack integrates 128GB of SRAM and 12TB of DDR5 memory. These specifications target AI models with trillions of parameters. The system aims to handle massive computational workloads efficiently.
Global availability for the Groq 3 LPX begins in early Q3 2026. Full production ramp-up is expected to start by the end of Q3 2026. Foxconn plans to supply 6,000 units of the Groq 3 LPX rack this year. The company expects to add another 10,000 units to production by 2027.
New accelerator targets trillion parameter AI models
Foxconn's market share in this sector is projected to rise from 55% to 60% in the second half of 2026. This growth follows rapid demand for the Vera Rubin platform and the Groq 3 LPX rack. Major customers for the Vera Rubin NVL72 rack include Google, Amazon AWS, and Microsoft. Production of the Vera Rubin VR200 NVL72 server will also begin in late Q3 2026.
Foxconn CEO Liu Yangwei stated the company currently produces over 1,000 accelerator boards. Foxconn plans to expand its manufacturing capacity to 2,000 boards by the end of 2026. The company aims to actively participate in the AI inference market alongside NVIDIA.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.