Back to today’s briefing

Story perspectives

NVIDIA's Blackwell Platform Boosts AI Efficiency Tenfold

2/13/2026

40 2 Full Breakdown

Summary

Story summary

NVIDIA's Blackwell platform has achieved a ten-fold improvement in token optimization for AI inference workloads over the Hopper generation.
The GB200 NVL72 system enables Baseten and DeepInfra to reduce costs per token by up to 10x.
During a recent launch, the platform processed 1.8 million users in 24 hours while maintaining low latency.
The upcoming NVIDIA Rubin platform is planned to further enhance performance and efficiency in tokenomics.

1 of 1

Story summary

NVIDIA's Blackwell platform has achieved a ten-fold improvement in token optimization for AI inference workloads over the Hopper generation.
The GB200 NVL72 system enables Baseten and DeepInfra to reduce costs per token by up to 10x.
During a recent launch, the platform processed 1.8 million users in 24 hours while maintaining low latency.
The upcoming NVIDIA Rubin platform is planned to further enhance performance and efficiency in tokenomics.