NVIDIA announced the Vera Rubin NVL72 system extended with the new Groq 3 LPX inference accelerator, delivering 3,400 ou

NVIDIA has announced that the Vera Rubin rack-scale system, paired with the newly released NVIDIA Groq 3 LPX, is now in full production. In testing with Gemma 4 31B, an open source agentic model, the Groq 3 LPX delivered 3,400 output tokens per second for 100,000-token long-context use cases that are critical to agentic systems, achieving performance 4x faster than the nearest alternative platform.

The shift in AI infrastructure from model training to inference has become the central focus of the industry. As agentic systems mature, they demand infrastructure optimized for throughput, responsiveness and economics at unprecedented scale. These systems generate more tokens, process dramatically larger context windows and increasingly collaborate with other AI systems to solve complex problems. NVIDIA is responding with what it calls extreme codesign—architecting compute, networking and inference acceleration as a unified system purpose-built for long-context inference and multi-agent applications.

The Vera Rubin platform serves as the foundation for this architecture. Vera Rubin is engineered to accelerate inference as agents reason over increasingly long sequences. The NVIDIA Spectrum-X Ethernet network moves massive data flows efficiently across AI factories. The Groq 3 LPX, a new low-latency inference architecture, is built to generate tokens at ultrafast speeds. Unlike standalone accelerators, Groq 3 LPX combines GPUs and specialized inference processors through extreme codesign, enabling Rubin GPUs to handle large-scale context processing while LPX accelerates latency-sensitive decode workloads. At scale, a rack-scale deployment can include 256 LP30 accelerators connected through direct chip-to-chip links, creating a highly efficient inference engine.

NVIDIA's partners are already adopting the Vera Rubin platform. Nebius is the first AI cloud to deploy the Groq 3 LPX, giving developers access to leading token generation speeds for highly responsive agentic AI applications. CoreWeave is deploying Spectrum-X Multiplane in production, connecting multiple Vera Rubin racks using parallel switches to provide high-bandwidth, flat and lossless AI networks. SpaceXAI plans to build its future AI architecture around NVIDIA Vera CPUs, deploying them to accelerate the CPU-intensive work behind agentic AI, including orchestration, tool use, code execution, data processing and simulation—with plans extending from data centers on Earth to orbital satellites.