Nvidia names its data-centre platforms after scientists, and each one has arrived bigger than the last. At GTC on March 16, 2026, it introduced Vera Rubin, named after the astronomer whose work provided evidence for dark matter, as the successor to Blackwell.

What Nvidia announced

The striking thing about Vera Rubin is that it isn’t a GPU. It is a platform of seven chips that Nvidia says are designed to work as one system, which it calls an AI factory:

  • Rubin GPU: the accelerator at the centre.
  • Vera CPU: Nvidia’s own Arm-based processor.
  • NVLink 6: the interconnect that ties GPUs together into one large machine.
  • ConnectX-9 and Spectrum-6: networking for scaling out across racks.
  • BlueField-4: a data processing unit that offloads storage, networking and security.
  • Groq 3 LPU: a processor focused on fast inference, from the Groq technology Nvidia brought in-house.

Nvidia said the chips were in full production and would ship in five rack designs, with partner systems due in the second half of the year. Its headline claim was up to 10 times the inference throughput per watt of Blackwell. As with any vendor benchmark, that figure depends heavily on the model and configuration.

The Computex update

At Computex in Taipei on May 31, Nvidia said Vera Rubin was ramping into full production, with production shipments starting in the autumn of 2026.

Why it matters

  • Inference is now the target. The emphasis on throughput per watt, and the inclusion of a dedicated inference chip, reflects where demand has moved: from training a model once to serving it billions of times. Power, not chips, is often the limiting factor in new data centres.
  • Nvidia is selling systems, not parts. With its own CPU, networking, DPU and interconnect, Nvidia controls more of each rack. That makes it harder for customers to mix vendors, and is part of why its margins have stayed around 75%.
  • What developers will notice: cheaper and faster inference reaching cloud providers over 2027, as Rubin capacity comes online. Model providers’ prices tend to follow the hardware.

Sources

  1. Nvidia: NVIDIA Vera Rubin opens agentic AI frontier (Mar 16, 2026)
  2. Nvidia: Vera Rubin in full production (May 31, 2026)

We link every claim to the reporting or official statement it comes from. Where a detail rests on one outlet’s reporting, we say so. Spotted an error? Reply to any newsletter and we’ll correct it.