Nvidia plans to integrate its Groq 3 LPX accelerator technology into the Vera Rubin platform in the second half of 2026, extending Groq’s role in Nvidia’s inference strategy after Nvidia licensed Groq’s chip architecture and hired its leadership without acquiring the company outright. Groq 3 LPX uses deterministic execution, scheduling computation and data movement cycle by cycle rather than relying primarily on reactive mechanisms such as caches and branch predictors. Nvidia says the design can deliver up to 35 times higher throughput per megawatt than prior-generation systems on models exceeding two trillion parameters. Preemptive Power and Clock Period Synthesis technologies cut voltage drops by more than 60% in internal testing, supporting a high-single-digit reduction in baseline voltage requirements and potentially double-digit power savings. Vera Rubin’s Intelligent Power Smoothing and DSX MaxLPS systems are designed to provision up to 40% more GPUs within the same power envelope while delivering 35% higher token throughput. The technology could help Groq’s cloud business, which operates 13 data centers, serves 6 million developers and thousands of companies, has $1 billion in fresh capital and is valued at $3.5 billion, down from $6.9 billion before the Nvidia transaction. Groq plans to quadruple capacity next year while competing with Fireworks, Together AI, CoreWeave, Amazon and Google. The Justice Department is reportedly examining the structure of Nvidia’s arrangement with Groq, and Nvidia has denied plans for China-specific versions of its Groq-based LPUs. Nvidia shares traded at $212.07 on September 15, 2026, giving the company a $5.15 trillion market capitalization.