AI, semiconductors, software, and big tech
New rack-scale system delivers over 1,000 tokens per second for trillion-parameter models, with modular architecture reducing datacenter setup from days to hours.