NVIDIA Vera Rubin NVL72 Delivers 30x Efficiency Boost for AI
NVIDIA unveiled performance data showing its Vera Rubin NVL72 platform delivers up to 30x higher throughput per megawatt and slashes token costs by 35x compared to its predecessor, the GB300 NVL72. This leap in efficiency is critical as agentic AI workloads—complex tasks involving multiple sub-agents and long-context handling—gain traction in industries ranging from finance to life sciences.
Agentic AI workloads differ significantly from traditional chat-based AI. These tasks often require hundreds of thousands of tokens in cumulative context, driving up computational demand and costs. NVIDIA has designed the Vera Rubin NVL72 to address these challenges by optimizing inference efficiency. According to NVIDIA’s tests using the SemiAnalysis AgentX workload, power-constrained AI factories can process 30x more work for the same energy cost, representing a dramatic improvement in operational scalability.
Why This Matters: AI Factory Economics
For AI factories, throughput per megawatt directly impacts output, while token cost defines profit margins. NVIDIA’s Vera Rubin NVL72 offers up to 40% more GPUs within the same energy budget, thanks to its DSX MaxLPS power management technology. This positions it as a critical tool for companies aiming to scale AI operations without ballooning costs.
NVIDIA has already announced partnerships with major cloud providers, including Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure, to deploy the Vera Rubin NVL72 platform. Companies like Bristol Myers Squibb are leveraging it to build advanced AI solutions in sectors like life sciences.
Performance Breakdown
The Vera Rubin NVL72’s efficiency gains stem from extreme hardware-software codesign. Features like distributed KV-caching, large-scale expert parallelism, and the latest NVLink 6 interconnect enable seamless GPU communication and memory management. Additionally, the platform’s Tensor Cores and Transformer Engine accelerate both context processing and response generation, critical to agentic AI performance.
Compared to the GB300 NVL72, which already outperformed the NVIDIA Hopper architecture by 15x in certain workloads, the Vera Rubin NVL72 raises the bar further. Early tests indicate up to 30x higher throughput on models like DeepSeek V4 Pro, although these results are pending independent review by SemiAnalysis.
Market Implications
NVIDIA’s focus on inference efficiency comes at a time when AI workloads are increasingly energy-intensive. With the global AI market projected to reach $1.8 trillion by 2030, platforms like Vera Rubin NVL72 could redefine the economics of AI deployment. Lower token costs and higher energy efficiency make it particularly attractive for enterprises operating large-scale AI infrastructures.
As of August 24, 2026, NVIDIA’s stock price sits at $210.14, down 2.13% over the last 24 hours. However, the long-term outlook appears strong, bolstered by the company’s leadership in AI hardware and its push into agentic workloads. Investors and industry stakeholders will likely watch for further updates on commercial adoption and additional performance benchmarks.
With Vera Rubin NVL72 now in full production, NVIDIA seems poised to maintain its dominance in the AI hardware market. The platform not only addresses the computational demands of next-generation workloads but also sets a new standard for energy efficiency—an increasingly critical factor as AI adoption scales globally.