Nvidia–Groq $20B Deal
Nvidia announced a $20 billion strategic tie-up with Groq at GTC to fold advanced inference into its Vera Rubin stack — a move positioned to lock in AI compute for high‑performance trading and DeFi models. The deal is framed as an ASIC‑blocking play that will shape next‑gen inference costs for AI trading infra. (digitimes.com)
The $20 billion transaction was struck on December 24, 2025 and was reported as a cash licensing-and‑acquihire arrangement for Groq’s assets rather than a straightforward corporate takeover. (cnbc.com) Nvidia’s first hardware product from the deal, the Groq 3 LPU (LPX) announced at GTC 2026, uses an SRAM‑first design with 512 MB of on‑die SRAM per die and a quoted 150 TB/s of memory bandwidth per die. (tomshardware.com) A full LPX rack is specified to house 256 LPUs for 128 GB of on‑chip SRAM and roughly 40 PB/s aggregate bandwidth, with Nvidia targeting Q3 2026 shipment and Samsung 4 nm manufacturing for the LPX modules. (tomshardware.com) Nvidia enumerated Vera Rubin platform components that now include NVL72 GPU racks, Vera CPU racks and dedicated Groq 3 LPX inference accelerator racks as part of a seven‑chip, five‑rack architecture positioned for large‑context, low‑latency inference. (nvidianews.nvidia.com) Groq had raised $750 million in September 2025 at a reported $6.9 billion post‑money valuation, and analysts and industry write‑ups note Nvidia’s $20 billion payment represented a multiple notably above that recent valuation. (groq.com) The deal was structured as a non‑exclusive technology license and broad hiring initiative—Groq’s cloud inference business remains separate while key engineers including founder Jonathan Ross and president Sunny Madra moved into Nvidia roles. (tomshardware.com) Nvidia published concrete throughput and cost targets tied to the integration—claiming LPX+Rubin configurations can deliver up to 35× higher throughput per megawatt for trillion‑parameter models and a target economics figure of about $45 per million tokens—while corporate guidance raised Nvidia’s long‑term revenue ambition toward $1 trillion by 2027, underlining the company’s inference‑driven financial thesis. (tomshardware.com)
