Nvidia just confirmed what the AI hardware world has been waiting to hear: the Vera Rubin platform is in full production and on schedule for partner availability in the second half of 2026. CEO Jensen Huang delivered the update at GTC 2026, positioning the architecture as the company’s most ambitious leap yet in the race to power agentic AI, foundation models, and memory-hungry inference workloads.
For anyone building, investing in, or simply watching the AI infrastructure buildout, this is the starting gun for the next hardware cycle. And for crypto markets, the downstream effects may be more significant than they first appear.
What Vera Rubin actually brings to the table
The flagship configuration is the NVL72 system. It packs 72 Rubin GPUs and 36 Vera CPUs into a single rack. The result: 3.6 exaflops of NVFP4 inference compute and 2.5 exaflops of training compute. In English: this is a machine that can run the largest AI models on the planet with headroom to spare.
Scale it up and the numbers get genuinely absurd. A full Vera Rubin POD can stretch to 40 racks, totaling 1,152 Rubin GPUs and roughly 60 exaflops of NVFP4 compute. To put that in perspective, the entire world’s supercomputing capacity was measured in single-digit exaflops just a few years ago.
