SAN JOSE, California, March 18, 2026 — Nvidia unveiled the Groq-3 AI accelerator and a new CPU server platform at its GTC 2026 conference, positioning the products as direct competitors to Intel’s Xeon and Gaudi lines in enterprise AI training and inference workloads.
The Groq-3 chip, the third generation of Nvidia’s custom inference accelerator, features a 3nm-class process node, 2.5x higher memory bandwidth than Groq-2, and up to 4x better performance per watt on large language model inference tasks.
Nvidia claims Groq-3 delivers sub-millisecond latency for real-time AI applications, making it suitable for edge deployment, customer service bots, and autonomous systems.
Alongside Groq-3, Nvidia introduced the Grace Vera CPU server—a rack-scale system pairing its Arm-based Grace CPU with Vera accelerators and high-speed NVLink interconnects.
The platform targets hybrid AI workloads that combine training and inference, offering up to 40% better energy efficiency than comparable Intel Xeon-based servers, according to Nvidia benchmarks.
CEO Jensen Huang positioned the announcements as a direct challenge to Intel’s enterprise dominance. “We’re bringing the same level of innovation that disrupted GPUs to the CPU and inference markets,” Huang said during his keynote.
Nvidia highlighted partnerships with Dell, HPE, and Lenovo to bring Groq-3 and Grace Vera systems to market in the second half of 2026.
The launches follow Nvidia’s $20 billion licensing and talent deal with Groq in late 2025, which brought Groq’s low-latency inference expertise in-house. Groq-3 incorporates elements of that technology, particularly deterministic execution and optimized memory access patterns.
Intel has not yet responded publicly. The company recently launched its Gaudi 3 accelerator and is preparing Xeon 6 processors for 2026, both aimed at AI and high-performance computing markets.
Analysts view Nvidia’s move as an aggressive expansion beyond its GPU stronghold, targeting Intel’s traditional enterprise strongholds. The Groq-3 and Grace Vera platforms are expected to pressure Intel’s pricing and market share in AI servers, where Nvidia already holds a commanding position.
