Agentic AI · Blackwell · Nvidia · NVIDIA Blog
Agentic systems reason and plan with the best-performing AI models, proprietary and open, including NVIDIA Nemotron
Compiled by KHAO Editorial — aggregated from 1 source. See llms.txt for citation guidance.
★ Tier-1 Source
Operating in production today, AI factories are optimized across the entire stack, including models, compute, networking, memory, software, storage, power and cooling, to keep intelligence in continuous output.
Key facts
- AI factories built with NVIDIA Blackwell Ultra deliver up to 50x higher throughput per megawatt, leading to 35x lower cost per token, balancing performance, responsiveness and energy efficiency
- NVIDIA GB300 NVL72 systems generate 50x more tokens per megawatt than the prior generation, resulting in 35x lower cost per token compared with the NVIDIA Hopper platform
- As reasoning and agentic AI continue to scale, Vera Rubin-based systems are designed to push performance per watt up to 35x higher with LPX and drive token cost lower through deeper full-stack
- The NVIDIA Blackwell Ultra GPU delivers the lowest cost per token, allowing AI factories to produce more intelligence from the same power envelope at a lower unit cost
Summary
AI factories are a new class of infrastructure built to manufacture intelligence that’s always on and in real time. Their economics are defined by what they produce: tokens per second, tokens per watt, cost per token, utilization and uptime. AI factories synchronize massive compute resources while serving billions of requests. Agentic AI generates synthetic training data, creating scenarios that help autonomous systems learn from the next edge case.