BREAKING
NVIDIA Pushes 'Cost Per Token' Metric
0
$
per 1M tokens (Blackwell)
0
$
per 1M tokens (Hopper)
0
x
cost reduction
Tokens Per Second, Per GPU
Blackwell
6000
Hopper
90
0
x
perf uplift in a month
0
x
peak throughput gain
Partners Report Early Gains
Baseten
reasoning
●
Serves DeepSeek V4 Pro on Blackwell
●
Up to 50% more tokens per second
Hippocratic AI
scale
●
30% throughput lift
●
Sub-half-second response, 10M calls
Can Rivals Close the Gap
AI NEWS BLITZ
NVIDIA says the key AI metric is now cost per token, not raw chip specs.