GTC 2026: Nvidia launches the Vera Rubin platform

2026-03-16 · Publisher : Nvidia · Domain : Artificial Intelligence

At GTC in San Jose, Jensen Huang presents the Vera Rubin platform: seven chips in production, including the first LPU from Groq, designed for agentic AI. He cites 1 trillion USD in Blackwell and Vera Rubin orders through 2027.

On March 16, 2026, during the GTC keynote at the SAP Center in San Jose, Jensen Huang presents the Vera Rubin platform, the successor to Grace Blackwell, which Nvidia positions for agentic AI and large-scale inference. ### What happened - The platform brings together seven chips announced as in full production: the Vera CPU, the Rubin GPU, the NVLink 6 switch, the ConnectX-9 SuperNIC, the BlueField-4 DPU, the Spectrum-6 Ethernet switch and the new Groq 3 LPU. - Groq 3 is the first chip from Groq, whose assets Nvidia largely acquired in December 2025 for 20 billion USD, its largest deal according to CNBC. - Nvidia offers the platform as five racks: Vera Rubin NVL72, Vera CPU rack, Groq 3 LPX, BlueField-4 STX storage and Spectrum-6 SPX Ethernet. - Jensen Huang sums it up: "The agentic AI inflection point has arrived with Vera Rubin." ### Key figures (Nvidia claims) | Item | Value | |---|---| | Vera Rubin NVL72 | 72 Rubin GPUs, 36 Vera CPUs | | MoE model training | one-fourth the GPUs needed with Blackwell | | Inference throughput per watt | up to 10x Blackwell | | Cost per token | up to 10x lower | | Groq 3 LPX rack | 256 LPUs, 128 GB SRAM, 640 TB/s | | Rubin + LPX gain | up to 35x more throughput per megawatt | | Blackwell and Vera Rubin orders through 2027 | 1 trillion USD, according to Jensen Huang | ### Why it is a turning point Nvidia no longer sells chips but integrated compute centers, from rack to "POD". Adding a processor specialized in low-latency inference next to GPUs acknowledges the explosion in tokens generated by agents. Dario Amodei (Anthropic) and Sam Altman (OpenAI) are among the customers quoted in the announcement. ### Caveats Performance gains are Nvidia's figures. Vera Rubin systems are announced for late 2026 and the LPX rack for the second half: real efficiency remains to be measured in production. ### Key takeaways Vera Rubin sets the architecture of 2026-2027 AI factories. Nvidia already announces Vera Rubin Ultra and the Kyber architecture (144 GPUs per rack) for 2027.

Tags : IA, NVIDIA, Calcul, GPU, Infrastructure, Agents