Nvidia Follows Google's Playbook With $20 Billion Groq Bet
Key summary
Nvidia licensed Groq's inference technology for $20 billion last December; Groq unveiled the Groq 3 LPU at GTC 2026 in San Jose.
The agreement is Nvidia's largest ever and grants a perpetual license to Groq's patent portfolio and software stack.
Groq will continue to operate independently under new CEO Simon Edwards.
Nvidia places Groq technology at the center of its inference roadmap; inference requires different silicon and is bottlenecked by memory bandwidth.
The forthcoming Rubin GPU offers 288 GB of high‑bandwidth memory at 22 TB/s while the Groq 3 LPU has 500 MB on‑chip SRAM delivering 150 TB/s; the Vera Rubin NVL72 pairs 72 Rubin GPUs with a Groq 3 LPX rack of 256 LPUs, and Nvidia's Dynamo orchestrates workloads in real time with GPUs handling prefill and LPUs handling decode.
