Nvidia Announces That Groq 3 LPX Is in Full-Scale Production. What This Means for NVDA Stock.

Pathikrit Bose

Wed, August 26, 2026 at 5:23 PM GMT+3 5 min read

At a time when agentic AI is gaining rapid traction, inference is becoming crucial. The world's most valuable company and GPU leader, Nvidia (NVDA), has often been flagged as a laggard by many in inference. Not willing to cede ground, the Jensen Huang-led company decided to buy its way into the market with the $20 billion acquisition (though the companies said it was a licensing deal) of inference-focused AI infrastructure company Groq late last year.

Now, Nvidia recently revealed Groq 3 LPX, the rack-scale system built around Groq's LPU technology, is in full-scale production.

More News from Barchart

The news was taken positively by the market, as shares of Nvidia are up about 2% since the announcement. Ahead of its Q2 earnings, NVDA stock is up 13.8% on a YTD basis, outperforming the S&P 500 Index ($SPX).

www.barchart.com

Live With Groq 3 LPX

Groq 3 LPX is Nvidia's rack-scale system, similar to Vera Rubin. However, the similarities end there as Groq 3 LPX is specifically designed to make AI inference, especially token generation, extremely fast and predictable for interactive and agentic AI applications. While the Vera Rubin GPU is used for general workloads such as training, the Groq 3 language processing unit (LPU) is designed specifically for ultrafast token generation. LPX is the rack that puts 256 of these LPUs together and connects them with Rubin.

Collectively, these 256 LPUs provide 315 PFLOPS of FP8 inference compute, 128 GB of on-chip SRAM, or static random access memory, 40 PB/s of SRAM bandwidth, and 640 TB/s of scale-up bandwidth. Moreover, Nvidia said it achieved 3,400 output tokens/second on Gemma 4 31B with a 100,000-token context in Artificial Analysis testing, with Nebius (NBIS) becoming the first AI cloud to adopt it.

Nvidia also says the combined Groq and Vera Rubin arrangement can deliver up to 35 times more inference throughput per megawatt than a conventional GPU-focused system. The benchmark confirms a leadership position for this specific model and test condition, but it does not prove that LPX will be four times faster across every model, batch size, or customer workload.

Overall, in the LPX design, Vera Rubin GPUs will handle training, prefill, and general computing tasks, while the Groq LPUs will handle rapid token generation. This division of labor is useful for coding assistants, customer service agents, search tools, and autonomous AI agents because a delay between tokens can make an application feel slow even if the total output is strong. However, customer adoption will depend on whether the speed advantage is large enough to justify a specialized system and whether Nvidia can keep its promised deployment schedule.

Building on Q1 With Q2 In Sight

Nvidia sustained strong performance in its most recent quarter, delivering substantial growth, exceeding expectations on both the top and bottom lines, and raising its forward guidance.

Revenue advanced 85% year over year to $81.6 billion, driven primarily by the data center segment, which expanded 92% to $75.2 billion. For the second quarter, the company anticipates revenue of $91 billion at the midpoint, excluding any contribution from China, compared with analyst projections of $91.73 billion.

Earnings per share rose to $1.87, reflecting a 140% increase from the year-earlier period and surpassing the consensus estimate of $1.75. This result extended Nvidia's streak of consecutive earnings beats to nine quarters. Gross margins improved meaningfully, expanding to 75% from 60.8% in the prior year.

Cash generation stayed healthy, with net cash from operating activities increasing to $50.3 billion from $27.4 billion a year earlier. At the end of the fiscal first quarter, Nvidia held $13.2 billion in cash while carrying a limited short-term debt balance of $1 billion.

From a valuation standpoint, NVDA stands apart from several other Magnificent Seven companies with relatively moderate metrics. The forward P/E ratio of 24.4x is just above the sector median of 22.43x, while the forward P/S multiple of 12.76 times and P/CF of 23.90x are within the range of the sector medians of 3.39 times and 19.82 times, respectively.

Nvidia is set to report its Q2 2027 earnings today, after the market closes. Apart from the commentary on the wider AI ecosystem, analysts are also expecting color on memory costs and whether they will affect Nvidia. In terms of numbers, the Street is expecting the Santa Clara, California-based chip giant to report revenue of $92.4 billion and earnings of $2.10 per share. Data center revenue is projected to double again to $85.92 billion.

Analyst Opinion

Analysts have deemed NVDA stock to be a "Strong Buy" with a mean target price of $307.38. This indicates upside potential of about 44.1% from current levels. Out of 48 analysts covering the stock, 43 have a "Strong Buy" rating, three have a "Moderate Buy" rating, one has a "Hold" rating, and one has a "Strong Sell" rating.