与“tag:"AI Inference"”匹配的图钉

Qualcomm entered the data-center AI market with the AI200 and AI250 accelerator cards and liquid-cooled racks, with 768 GB of memory per card, due in 2026 and 2027; its stock jumped 11% that day.

Qualcomm will build customised AI inference silicon and up-to-1.6T optical connectivity for Amazon's data centres over several generations, and gave Amazon a warrant for up to 25 million Qualcomm shares vesting against up to $60 billion of purchases.

Qualcomm's second-generation data-center AI accelerator, the Dragonfly AI250, is due in 2027 with High Bandwidth Compute near-memory stacking that gives 133 TB/s of effective bandwidth per card, 18 times the AI200.