Particle.news
Download on the App Store

Nvidia Strikes ~$20 Billion Groq Deal to Power Its AI Inference Push

The move targets the sector’s shift to always-on inference workloads that prize low latency and efficiency.

Overview

  • Nvidia announced a definitive agreement to pay about $20 billion in cash to license Groq’s technology and hire the startup’s core engineering team to bolster real-time inference performance.
  • Shares rose roughly 1% to close at $190.53 following the news, lifting Nvidia’s market value to more than $4.6 trillion on a relatively light post-holiday trading day.
  • The company and analysts framed the transaction around the 2025 “Inference Flip,” where revenue from running AI models surpassed one-time training, making low-latency accelerators a priority.
  • Nvidia says it remains sold out of cloud GPUs and is working through a large order backlog, reinforcing demand visibility heading into 2026.
  • Competition is building from Chinese GPU newcomers and hyperscalers’ custom chips, yet firms such as Bernstein and Cantor reiterated bullish views on Nvidia’s 2026 setup, including an Outperform and a top-pick call.