Inferact says its vLLM optimizations on GB300 improve TCO

Inferact says its team has been optimizing vLLM, the open-source inference project, on Nvidia GB300 hardware, and that the gains translate directly into better total cost of ownership.

The company points to current results on the InferenceX benchmark site, which is run by SemiAnalysis, and says more optimizations are coming soon. The claim comes from Inferact's own X post thanking SemiAnalysis for a shoutout; the post gives no figures for the size of the improvement.