NVIDIA Rubin platform enters full production — AWS, Google Cloud, Microsoft, OCI line up as first H2 2026 deployment customers
NVIDIA confirmed Rubin platform full production this week, with Rubin-based systems shipping to AWS, Google Cloud, Microsoft, and Oracle Cloud Infrastructure as the first wave of cloud deployments through the second half of 2026. The company is now projecting at least $1 trillion in cumulative Blackwell + Vera Rubin demand through 2027 — double the prior forecast and a number that recasts datacenter capex assumptions for every hyperscaler.
The Rubin platform consolidates six new chip designs into a single supercomputer-class system targeting inference rather than training. NVIDIA's framing — "2026 is the year of inference" — is doing strategic work: it tells customers the training-cluster buildout of 2024–2025 is being followed by an inference-cluster buildout of comparable scale, and that the two are not substitutes.
The $1T cumulative-demand figure is the more consequential number. It implies a hyperscaler capex curve that runs through 2027 without a softening point. Meta's pivot to fund a $115–135B AI infrastructure spend via layoffs — and the comparable signaling from Microsoft and Oracle — looks less like a single-year reallocation and more like the new baseline. Anyone modeling AI infrastructure as a cyclical industry has to revisit the cycle assumption.
NVIDIA — Rubin Platform AI Supercomputer → · CNBC — Nvidia GTC AI chip pivot CPU center stage → · S&P Global — AMD next-generation AI chips power 2026 data center growth →