Inference Surge & Partners
Jensen Huang framed a coming inference boom — he cited a 1,000,000× jump in deployed compute and a 350× token‑generation leap tied to agent platforms and enterprise AI factories. NVIDIA also rolled partnerships to put AI factories on NVIDIA Ethernet/InfiniBand with players like Netris and Spectro Cloud to industrialize deployment (x.com) (x.com) (youtube.com).
Jensen Huang delivered the GTC keynote in San Jose on March 16, 2026, and used the stage to outline an “AI factory” roadmap while projecting at least $1 trillion in AI infrastructure demand through 2027. (nvidia.com) NVIDIA showed its Vera Rubin family (NVL72) on stage as a multi‑chip, multi‑rack architecture described in the keynote as a 7‑chip, 5‑rack system, and vendor previews cite up to 5× inference throughput and a promise of ~10× lower cost‑per‑token versus Blackwell with availability targeted in the second half of 2026. (youtube.com) NVIDIA also unveiled DSX Air, a cloud SaaS simulation platform to model full AI‑factory stacks end‑to‑end before hardware ships, with NVIDIA naming CoreWeave and other partners as early users for pre‑deployment validation. (blogs.nvidia.com) Netris announced integration and immediate support for NVIDIA DSX Air on March 16, 2026, saying the partnership brings Netris’ network automation and multi‑tenancy capabilities into DSX Air so customers can validate networking and tenant isolation in simulation prior to rollout. (netris.io) Spectro Cloud published PaletteAI general availability and a validated AI‑factory blueprint with Netris at GTC 2026, describing automated multi‑fabric networking and hardware‑enforced tenant isolation as part of a deployable stack from bare metal to model deployment in a March 17, 2026 release. (netris.io) NVIDIA pushed networking as a core pillar for these factories with Spectrum‑X (Ethernet) and Quantum‑X (InfiniBand) photonics switches featuring co‑packaged optics and published configurations that scale into the 100–400Tb/s class while promising multi‑fold power and density gains for hyperscale AI fabrics. (nvidianews.nvidia.com)
