NVIDIA Enters Full Production of Groq 3 LPX AI Inference Accelerator Chips, Supercharging Vera Rubin With The Fastest Token Generation Speeds Ever Recorded

Nvidia is investing $6 billion to license AI technology from Poolside and hiring its engineers to develop an open-weight AI model, competing with Chinese rivals like DeepSeek and Moonshot AI. Nvidia will also invest $1 billion in Poolside at a $12 billion valuation. Analysts suggest this move is a strategic response to competitive pressure from Chinese AI models, which have gained significant market share.

Nvidia (NVDA) is in talks to invest in Perplexity at a valuation above $30B, a 50% increase from its prior $20B valuation. Perplexity's revenue grew from under $250M to over $750M this year, driven by its Perplexity Computer product. The company plans a 2028 IPO, backed by Nvidia, Jeff Bezos, and SoftBank.

Nvidia began full production of Groq 3 LPX AI racks, acquired from startup Groq for $20B. The systems, designed for AI inference, will deploy at cloud provider Nebius. Each rack has 256 chips, generating 3,400 tokens/second. Nvidia aims to complement, not replace, GPUs with Groq tech. CEO Jensen Huang projects $1T in sales from Blackwell and Vera Rubin platforms by 2027. AMD is also entering the AI inference hardware market.

NVIDIA's AVO software improved Anthropic's Claude Opus 5 AI model to achieve a perfect score on the ARC-AGI-3 benchmark without retraining. AVO manages memory, plans tasks, and corrects errors, enhancing performance. The benchmark tests learning and adaptability in game-like environments. NVIDIA aims to make AI more reliable for enterprise use, potentially benefiting companies like Microsoft and Salesforce, as well as itself.

Nvidia agreed to pay $6 billion to license Poolside's AI software and invest $1 billion in the startup, acquiring over 100 engineers. Poolside's Model Factory software and talent will join Nvidia's Nemotron open-weight models. Nvidia previously executed a similar deal with Groq. Poolside's failure to secure $2 billion in funding highlights the importance of compute resources in the AI sector.

Microsoft received the first production NVIDIA Vera Rubin systems for its AI data centers. CEO Satya Nadella confirmed the deployment, highlighting the milestone ahead of NVIDIA's earnings report. Each NVL72 rack costs $7M-$8M, features 72 Rubin GPUs, and delivers 3.6 exaflops of performance. NVIDIA's largest cloud customers, including AWS and Google Cloud, are part of the initial deployment cohort. Analysts note the transition to rack-scale systems like Vera Rubin could drive significant AI inf