NVIDIA’s next platform is real and shipping — but the first allocation is already spoken for. For everyone not named CoreWeave, Lambda, Oracle or Microsoft, Blackwell remains the capacity you can actually deploy.
NVIDIA announced at GTC Taipei that the Vera Rubin platform has ramped into full production, with production shipments beginning this fall. The platform combines Vera Rubin NVL72 GPUs, Vera CPUs, BlueField-4 storage processors and Spectrum-6 Ethernet into a rack-scale system NVIDIA positions for agentic AI workloads, claiming roughly 10× agent throughput at scale versus the Grace Blackwell generation.
The named early adopters tell the real story: CoreWeave, Lambda, Oracle Cloud Infrastructure and Microsoft Azure, supported by an OEM bench of Dell, HPE, Lenovo and Supermicro. As with every generation before it, first-wave Rubin allocation flows to the largest committed buyers — and the manufacturing ecosystem behind it spans 350+ factories, all of which hyperscale orders will keep busy well into 2027.
For the rest of the market, the practical question isn’t when Rubin ships — it’s when Rubin ships to you. History from the H100 and Blackwell ramps says that gap is measured in quarters, not weeks.
What it means if you’re buying: Rubin’s arrival doesn’t change your 2026–2027 build math — allocation does. B300-class capacity remains the compute you can contract today and stand up this year, and a 36-month B300 term signed now runs largely inside the window before Rubin reaches general availability at scale.
What to do about it: On the marketplace right now: MAGNETAR (1,024× B300, Ohio, December 1 RFS, from $4.75/GPU-hr), BLAZAR (1,024× B300 build-to-order, Washington, 10-week delivery) and ALCOR (up to 4K B300 GPUs, India). Capacity with a date on it beats a roadmap with a queue on it.
Original announcement · nvidianews.nvidia.com ↗ · Summary and analysis by Ax3
