NVIDIA (NVDA)’s $20 Billion Groq Bet Goes Live This Year With New AI Racks

CNBC reported that NVIDIA Corporation (NASDAQ: NVDA )'s Groq 3 LPX rack is in full production and will be deployed alongside its Vera CPUs and Rubin GPUs at neocloud Nebius later this year, Nvidia senior director Dion Harris told reporters.

The rack marks the commercialization of technology from Nvidia's $20 billion purchase of Groq's assets in December, its largest deal on record. Each liquid-cooled rack packages 256 Groq chips. Nvidia said the rack can deliver 3,400 tokens per second, noting a benchmark from Artificial Analysis. Groq's chips are manufactured by Samsung, unlike Nvidia's own GPUs, which are made by Taiwan Semiconductor Manufacturing Co.

The $20 billion Groq deal represents a manageable financial risk for NVIDIA Corporation (NASDAQ: NVDA ) relative to the strategic capabilities it could add to the company. 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. That gives Nvidia the flexibility to make large strategic investments while expanding into new AI chip technologies without materially straining its balance sheet.

Groq's technology extends Nvidia's dominance across the full AI compute stack rather than creating a rival product line. LPX handles low-latency inference while Nvidia's GPUs continue handling training and large-context processing. The rack can pair with Vera Rubin chips without customers changing their CUDA workflows, deepening the software lock-in that has made Nvidia hard to displace.

Execution has been fast, which matters in a market moving this quickly. Nvidia went from announcing the Groq purchase in December to full production and a named customer, Nebius, in eight months. It is a pace that shows Nvidia can absorb acquired technology and ship it rather than let it stall in integration.

Nvidia's inference speed now has a quantified edge over a key rival's approach. The Groq rack's 3,400 tokens per second compares with the 750 tokens per second OpenAI has promised for its Cerebras-powered "Ultrafast" mode. It gives Nvidia a benchmark it can point to as cloud providers shop for faster, more responsive AI inference.

NVIDIA Corporation (NASDAQ: NVDA ) paid $20 billion for Groq's technology and talent without acquiring the company itself. Since Nvidia licensed Groq's technology and hired its employees rather than purchasing Groq outright, investors have less visibility into the full value of what Nvidia obtained and whether the transaction can generate returns proportionate to its unusually large price tag.