GTC: Rubin + Blackwell leap
NVIDIA’s GTC buzz centered on the Vera Rubin architecture and claims of 3–4x Blackwell density, CG‑HBM memory and roughly 10x perf/watt gains — the keynote also teased Blackwell Ultra and Nemotron models. ( ) Speakers flagged massive order flow — about $1 trillion of Blackwell/Rubin orders through 2027 — and Jensen Huang projected quarterly revenue of $78B (+77% YoY) as customers push enterprise AI deployments. ( )
The Rubin superchip shown onstage combines two Rubin GPU dice with an 88‑core Arm Vera CPU and ships with up to 288 GB of HBM4 on‑package, a configuration NVIDIA and partners say drives much larger memory pools per socket than prior generations. (hothardware.com) NVIDIA described the Vera Rubin platform as a codesigned stack of seven chips — Rubin GPUs, Vera CPUs, NVLink‑6 switch, ConnectX‑9 SuperNIC, BlueField‑4 DPU, Groq 3 LPX inference LPUs and Spectrum‑6 switches — intended to operate as rack‑scale “AI factory” systems. (developer.nvidia.com) Company materials and keynote slides quantify system‑level efficiency gains rather than single‑die benchmarks, citing up to 35× more tokens‑per‑watt in some Rubin POD configurations and up to 10× inference perf‑per‑watt improvements at rack scale compared with earlier platforms. (developer.nvidia.com) NVIDIA said demand translated into roughly $1 trillion of booked or prospective orders for Blackwell and Rubin capacity through 2027, a figure Jensen Huang repeated during the GTC keynote and which analysts and outlets flagged as orders across hyperscalers and cloud providers. (cnbc.com) The company paired that rigour with financial guidance: NVIDIA reported a $68.1 billion quarter and issued guidance of about $78 billion for the following quarter (a roughly +77% year‑over‑year increase in the guide), figures that management and market coverage tied to accelerating enterprise AI deployments. (finance.yahoo.com) NVIDIA says Rubin silicon is already in full production and that Rubin‑based instances will reach major cloud partners and specialist providers in the second half of 2026, with roadmap slides and partner lists naming AWS, Google Cloud, Microsoft, Oracle, CoreWeave and others as early deployment targets. (investor.nvidia.com)
