For companies running AI services, the hardest problem is no longer simply building a powerful model. It is keeping thousands of users supplied with fast answers while controlling the cost of the chips, electricity, cooling and networking required to serve them.
That is why a post from SemiAnalysis on X has attracted attention. The analyst firm said a Vera Rubin NVL72 system delivers approximately 67 times the throughput per dollar of Nvidia’s GB300 platform when measured against a service level objective of 170 tokens per second.
In practical terms, the comparison appears to ask how much response capacity an operator can buy while maintaining a defined speed for users. That is a more commercially meaningful measure than a peak benchmark completed under ideal conditions. If accurate, the result would suggest that Vera Rubin could sharply reduce the cost of running demanding AI assistants and agentic applications.
The post offered little detail about how the number was calculated. It did not disclose the hardware configurations, prices, model, workload, power costs or software assumptions behind the comparison. Each of those variables can substantially change the outcome. A result based on one model or serving pattern may not represent the economics of every AI deployment.
SemiAnalysis also said Nvidia executive Ian Buck would present “real AgentX” at the AI Infra Summit in Santa Clara. That presentation could provide a clearer view of the software and hardware used to support the claim. It may also explain how Nvidia expects AI agents, which perform multiple steps rather than answer a single prompt, to be served at predictable speeds.
Why the claim matters
A major improvement in throughput per dollar would affect more than Nvidia’s product rankings. Cloud providers could reconsider expansion plans, while enterprises might bring larger workloads in house. Lower serving costs could also make persistent AI assistants more affordable for developers and consumers.
For now, however, the 67 times figure is a reported claim, not an established industry result. Until the methodology is published or Nvidia confirms the comparison, buyers will need to treat it as an intriguing signal rather than a procurement decision.
Daniel Reyes is not a person. No notebook, no deadlines, no face behind the name — just a byline this newsroom publishes under. Here is the production line underneath it, because a name beside a portrait reads like a journalist, and this one is not one.
The models. Writing: gpt-5.6-luna and qwen3-max. Out on the live web: gpt-5.6-luna and gpt-5.6-terra. Pictures: gpt-image-1 and gpt-image-1-mini. Swap one in the newsroom and this line swaps with it — it is read off the machines, not typed here.
How a story is made
- Research. The searching model reads around the story, pointed at primary sources — the filing, the post, the repository — rather than at somebody else's write-up of them.
- Writing. The writing model drafts it against what was found, at Daniel Reyes's usual length and in Daniel Reyes's usual register.
- The loop. A reviewer reads the draft and sends it back with notes. Then reads it again. A piece can go round several times before it leaves the building.
- Enrichment. A quotation has to appear word for word on the page it is taken from. A chart may only use figures that appear in the source it cites. Whatever fails is dropped, and the reason is kept.
- Fact check. A last pass hunts for claims the article makes and its sources do not.
- A human stop. Sensitive subjects are held for a person to read before publication, and a person can kill any of it at any point.
If that sounds less like a newsroom and more like a factory: quite. It is called Press Factory.
This article was generated using AI and published automatically without human pre-publication review.
How this article was made
The article was produced by the Grandmonts Media News Engine using automated research, drafting and verification workflows. No human editor reviewed the article before publication. Grandmonts Media remains responsible for the published content. Errors can be reported at office@grandmonts.cz.