NVIDIA's New Vera Rubin Chip Does 30 Times More AI Work Per Watt Than Its Predecessor
Fresh benchmark data puts NVIDIA's Vera Rubin NVL72 far ahead on energy efficiency for the complex, multi-step AI tasks that companies are building into their products at speed.
Key points
- NVIDIA's Vera Rubin NVL72 delivers up to 30 times higher AI throughput per megawatt than the GB300 NVL72, by NVIDIA's own measurements.
- Agentic AI workloads, where software carries out long chains of tasks automatically, consume 15 times more tokens than a simple chat request, per OpenRouter data.
- Token costs drop by up to 35 times on Vera Rubin NVL72 compared with GB300 NVL72.
- Results are pending independent review by SemiAnalysis and don't yet include performance from the Vera CPU component.
- Vera Rubin is in full production and shipping now.
What is an AI agent, and why does it eat so much power?
An AI agent is software that works through a long chain of steps on its own, rather than answering a single question. Think of it less like a search engine and more like a junior analyst who can open files, run other tools, and write up findings without being prompted at every stage.
That independence is expensive. According to OpenRouter, a service that routes traffic between AI models, agentic tasks use 15 times more tokens than a plain chat message. Tokens are the small chunks of text, roughly three-quarters of a word each, that AI models read and write. More tokens means more chips and more electricity.
When an agent researches a company for an investment report, it queries databases, reads filings, spins up a second agent to model valuations, and stitches everything together. Every step feeds its output into the next, so the text the model must hold in memory grows continuously. Some sessions run to hundreds of thousands of tokens.
What did NVIDIA actually announce?
The numbers are large. NVIDIA says its Vera Rubin NVL72 delivers up to 30 times more agentic work per megawatt than the GB300 NVL72, its current shipping platform. The 30x figure applies specifically to the DeepSeek V4 Pro model on the SemiAnalysis AgentX benchmark.
For context, the GB300 NVL72 was already 15 times more efficient than the earlier Hopper architecture on the same workloads.
| System | Throughput vs. Previous gen | Token cost vs. GB300 NVL72 |
|---|---|---|
| Hopper (previous gen) | baseline | not measured |
| GB300 NVL72 | 15x better per megawatt | baseline |
| Vera Rubin NVL72 | 30x better per megawatt | 35x lower |
The SemiAnalysis AgentX benchmark replays recorded real-world AI coding sessions, complete with the tool calls and sub-tasks a genuine agent would trigger. NVIDIA notes that SemiAnalysis is still reviewing the results, and that the figures don't yet account for the Vera CPU, which handles tool-calling tasks. Final numbers could move in either direction.
We covered the GB300's predecessor architecture in August 2026, when Nvidia announced its Groq 3 LPX production plans, the same partner ecosystem Vera Rubin now ships into.
What does this mean for ordinary people?
Most readers won't buy one of these systems. The impact arrives indirectly, through products built on top.
When the cost of running an AI agent drops sharply, businesses can afford to let those agents work longer and serve more customers without the economics falling apart. The AI assistant inside a customer-service app or an accounting tool can handle harder problems rather than handing off to a human the moment things get complicated.
For data centre operators, the electricity saving matters as much as raw speed. Power limits are strict, and 30 times more work per watt means fitting far more productive capacity into the same building. Worth watching: NVIDIA raised server prices by more than 15% for major customers just weeks ago, so cheaper inference per token doesn't automatically mean cheaper access to the hardware that delivers it.
Common questions
Are these numbers independently verified?
Not yet. NVIDIA measured these figures using the SemiAnalysis AgentX benchmark, and the results are pending SemiAnalysis review. The figures also leave out Vera CPU performance.
When can customers use Vera Rubin NVL72?
NVIDIA says Vera Rubin is in full production and scaling across its partner ecosystem now.
