Nvidia Corporation'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, which was the company's largest deal on record. Each liquid-cooled rack contains 256 Groq chips. According to a benchmark from Artificial Analysis, the rack can deliver 3,400 tokens per second. Unlike Nvidia's own GPUs, which are manufactured by Taiwan Semiconductor Manufacturing Co., Groq's chips are manufactured by Samsung.

Strategic Integration and Performance Benchmarks

The Groq technology is intended to extend Nvidia's dominance across the AI compute stack. While Nvidia's GPUs continue to handle training and large-context processing, the LPX technology handles low-latency inference. The rack can pair with Vera Rubin chips without requiring customers to change their CUDA workflows.

Nvidia's inference speed now possesses a quantified edge over a specific rival approach. The Groq rack's 3,400 tokens per second compares to the 750 tokens per second OpenAI has promised for its Cerebras-powered "Ultrafast" mode.

The execution of the deal has been rapid, moving from the announcement of the Groq purchase in December to full production and a confirmed customer, Nebius, in eight months. However, the LPX is a narrow, specialized product rather than a broad platform. The design of the chip makes it suited for inference, but brings limitations for other tasks such as training, meaning its addressable use case is smaller than Nvidia's core GPU business.

Financial Structure and Investment Portfolio

Nvidia paid $20 billion for Groq's technology and talent without acquiring the company itself. Because Nvidia licensed the technology and hired the employees instead of purchasing Groq outright, investors have less visibility into the full value obtained and whether the transaction will generate returns proportionate to the price tag.

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, providing flexibility to expand into new AI chip technologies without materially straining its balance sheet.

Nvidia is managing several simultaneous investments, including:

• Up to $100 billion committed to OpenAI
• $5 billion committed to Intel
• A $900 million licensing deal for Enfabrica's team
• Smaller sums invested in Crusoe, Cohere, and CoreWeave

As of Q2 of 2026, Insider Monkey's database shows that 285 hedge funds held NVIDIA Corporation, up from 275 in the first quarter. Rival AMD was held by 164 hedge funds.

Market Adoption and Perspectives

The rollout currently rests on one named customer, Nebius, which is the only cloud provider confirmed to deploy the Groq rack so far. It remains unclear whether the technology will see broad adoption or remain a niche addition.

Market perspectives on the move vary:

• Optimists believe Nvidia will succeed by moving quickly, outperforming competitor technology, and maintaining customer attachment to its CUDA software.
• Pessimists express concern that the company may struggle by allocating capital to too many projects simultaneously, purchasing technology rather than entire businesses, and relying on a product that currently has only one customer.