Nvidia's Groq 3 LPX inference rack has entered full production and will deploy at neocloud provider Nebius before year-end, eight months after Nvidia's $20 billion licensing deal for Groq's chip technology. The rack claims a speed edge over a rival system from OpenAI and Cerebras, but the rollout still rests on a single named customer.
Groq rack enters production for Nebius
Nvidia's Groq 3 LPX rack is in full production and will deploy alongside the company's Vera CPUs and Rubin GPUs at neocloud Nebius later this year, Nvidia senior director Dion Harris told CNBC. Each liquid-cooled rack packages 256 Groq chips, which are manufactured by Samsung rather than Nvidia's usual chipmaker, Taiwan Semiconductor Manufacturing Co. Nvidia confirmed full production status on August 24, 2026, with deployment planned before year-end through Nebius's Token Factory platform.
A licensing deal, not an acquisition
The rack commercializes technology from Nvidia's $20 billion deal for Groq's assets. The deal was signed on Christmas Eve 2025. Rather than buying Groq outright, Nvidia paid for a non-exclusive license to its LPU designs and hired select personnel, including Groq founder and CEO Jonathan Ross, while Groq continues operating independently under Simon Edwards. Groq has since raised $650 million in June 2026 and $350 million more in August 2026, reaching a $3.5 billion valuation, with Nvidia participating in the August round.
An inference-speed edge over a rival's numbers
Nvidia says the rack can deliver 3,400 tokens per second, citing a benchmark from Artificial Analysis. That compares with the 750 tokens per second OpenAI has promised for its Cerebras-powered "Ultrafast" mode. Nvidia separately touts up to 35 times the throughput per megawatt compared with conventional setups. Bernstein analyst Stacy Rasgon told CNBC that Nvidia is financially strong enough to absorb a deal of this size with little impact on its overall position, giving the company room to keep making large strategic bets without straining its balance sheet.
Reach outweighs the risk, for now
Nvidia is running several such bets simultaneously, having also committed up to $100 billion to OpenAI and $5 billion to Intel, plus smaller sums to Crusoe, Cohere, and CoreWeave. Nebius remains the only cloud provider confirmed to deploy the Groq rack so far, so it stays unclear whether the technology sees broad adoption or stays a niche addition. Insider Monkey's database shows NVIDIA was held by 285 hedge funds in Q2 2026, up from 275 in the first quarter, versus 164 for rival AMD.
Sources: Insider Monkey, Crypto Briefing
Trading involves risk.
Risk Disclaimer
