Nvidia’s Ultra-Low-Latency AI Inference LPX Racks Hit Full Production
- On Monday, August 24, 2026, Nvidia Corporation announced its Groq 3 LPX inference system entered full production, with deployment at neocloud Nebius scheduled for later this year.
- This milestone arrives eight months after Nvidia closed a $20 billion licensing agreement with chip designer Groq Inc on December 24, 2025. Groq remains an independent entity despite Nvidia hiring much of its engineering staff.
- Benchmarks by Artificial Analysis indicate the LPX system processes 3,400 tokens per second using Google's Gemma 4 31B model, which Nvidia claims is 4x faster than the nearest alternative platform.
- Nvidia is pairing the LPX with Vera Rubin processors to accelerate inference, estimating the combined system delivers up to 35 times the throughput per megawatt. "This isn't about replacing GPUs," Nvidia senior director Dion Harris said.
- As market focus shifts toward inference, the industry anticipates Nvidia's earnings report on Wednesday, August 26. Analysts expect the company's data center revenue to reach nearly $92 billion.
25 Articles
25 Articles
Nvidia announced that the Groq 3 LPX systems have entered serial production and will become operational by the end of this year, marking the first important commercialization of the technology taken by the company's largest acquisition in history, CNBC broadcasts.
NVIDIA has begun mass production of the GRACK3 LPX, a chip dedicated to AI inference, utilizing Samsung Electronics' Pyeongtaek Campus 4nm process as a production base. This chip, which has secured Nevius as its first customer, is characterized by maximizing AI responsiveness by integrating 256 language processing units.
(San Francisco = Yonhap News) Correspondent Kwon Young-jeon = NVIDIA's inference-dedicated chip, manufactured by Samsung Electronics, has secured customers and entered the full-scale mass production phase.
Nvidia’s $20 Billion Groq Bet Is Going Live Before the End of 2026. Here’s What It Means for Investors.
Nvidia says its Groq 3 LPX inference chip is now in production, with Nebius lined up as the launch customer before the year's end. The announcement caps an eight-month sprint since Nvidia's reported $20 billion Groq licensing deal.
Coverage Details
Bias Distribution
- 50% of the sources lean Right
Factuality
To view factuality data please Upgrade to Premium
