NVIDIA unveils Vera era
NVIDIA announced new Vera CPUs and Vera Rubin GPU platforms at GTC, pitching agentic-AI performance gains and landing hyperscaler customers like Alibaba, Meta and Oracle. The push comes with major OEM partnerships (Dell, HPE, Lenovo, Supermicro) — a clear signal hyperscalers are doubling down on vertically integrated GPU+CPU stacks.
Vera is an 88-core(winbuzzer.com) Arm CPU that NVIDIA says uses Olympus cores to deliver roughly a 1.5x IPC uplift and the company’s fastest single‑thread performance claims.(tomshardware.com) The Vera Rubin NVL72 configuration pairs 72 Rubin GPUs with 36 Vera CPUs over NVLink‑6 and, according to NVIDIA, can train mixture‑of‑experts models using one‑quarter the GPUs of Blackwell while delivering up to 10x higher inference throughput.(marketchameleon.com) NVIDIA stated seven new chips in the Vera Rubin family are already in full production to “scale the world’s largest AI factories,” a milestone the company emphasized in its GTC materials.(nvidianews.nvidia.com) Cloud and infrastructure partners named in NVIDIA’s release beyond the headline hyperscalers include CoreWeave, Lambda, Nebius and Nscale, with ByteDance also listed among early collaborators.(manilatimes.net) The supply‑chain roster expands past tier‑one OEMs to manufacturing partners such as ASUS, Compal, Foxconn, GIGABYTE, Pegatron, Quanta Cloud Technology (QCT), Wistron and Wiwynn in NVIDIA’s partner list.(marketchameleon.com) NVIDIA’s platform announcements bundled new rack SKUs — Vera CPU racks, NVL72 Rubin GPU racks, BlueField‑4 STX storage racks, and Spectrum‑6 SPX Ethernet fabric — plus third‑party Groq 3 LPX inference racks, with some systems slated for broader availability in the second half of 2026.(nvidianews.nvidia.com) Onstage at GTC, Jensen Huang projected demand large enough to approach $1 trillion in orders for Blackwell and Vera Rubin through 2027, a figure CNBC reported from the keynote.(cnbc.com)
