Nvidia’s $20B Groq Bet
Nvidia is making a roughly $20 billion strategic bet on Groq and its Groq 3 LPU, signaling a move beyond GPU‑only inference and toward heterogeneous AI compute architectures — a potential inflection for hardware roadmaps and specialized inference hiring. (forbes.com)
Nvidia struck a roughly $20 billion non‑exclusive licensing-and‑assets deal with Groq on December 24, 2025 that moves Groq’s engineering team and IP into Nvidia while leaving Groq’s cloud unit operating independently. (cnbc.com ). (cnbc.com) Nvidia unveiled the Groq 3 Language Processing Unit (LPU) at GTC on March 16, 2026 as the first product built from the licensed Groq technology and engineering collaboration. (tomshardware.com ). (tomshardware.com) Each Groq 3 LPU packs about 500 megabytes of on‑chip SRAM and delivers approximately 150 terabytes per second of SRAM bandwidth, and Nvidia’s LPX rack integrates 256 LPUs for rack‑scale inference. (nvidia.com ). (nvidia.com) Samsung Foundry confirmed mass production of the Groq 3 LPU on a 4‑nanometer process after Nvidia’s GTC announcements, making Samsung a primary foundry partner for the new inference accelerator. (dataconomy.com ). (dataconomy.com) Groq founder Jonathan Ross and president Sunny Madra are transitioning into Nvidia roles as part of the deal while Groq named former CFO Simon Edwards to lead the remaining Groq entity and continue GroqCloud operations. (groq.com ). (groq.com) Axios and other outlets report most Groq shareholders will receive staged payouts tied to the $20 billion arrangement—roughly 85% upfront, about 10% mid‑2026 and the remainder by year‑end 2026—while lawmakers have opened antitrust questions about the structure. (axios.com ) (axios.com)
