Qualcomm Demonstrates AI200 Inference Rack: Integrating Proprietary Accelerators with AMD CPUs

K-LinePoet · 2026-03-15T14:43:33+00:00

Qualcomm showcased the AI200 rack-scale AI inference solution at MWC26, which will be commercialized in the second half of the year. Each rack stands at 51U in height and contains 56 AI200 accelerator cards and 14 AMD EPYC processors, with total memory capacity reaching 43TB. The company also plans to introduce AI250 and AI300 systems in the future.

K-LinePoet

2026-03-15 14:43:33

Abstract generation in progress

IT House, March 3 — According to the official Qualcomm blog and a现场报道 by German media ComputerBase, Qualcomm showcased a physical prototype of its AI200 rack-mounted AI inference solution at MWC26 in Barcelona. The product is expected to be commercially available in the second half of this year.

IT House learned that each AI200 rack has a total height of 51U, consisting of 7 sets of 5U high systems. Each 5U system contains 4U for installing AI200 acceleration cards, with 2 cards deployed inside a single 1U tray; the remaining 1U is equipped with 2 AMD EPYC Milan processors. For connectivity, PCIe is used within small ranges, while 800G Ethernet is used for larger-scale connections.

Overall, a single AI200 rack includes 56 AI200 acceleration cards, with a total memory capacity of 43TB across all cards; additionally, the rack contains 14 AMD EPYC server processors.

According to German media reports, Qualcomm’s AI250 rack system in 2027 will still use AMD processors on the head node. The data center products introduced in 2028 will include the AI300 system and self-developed CPUs.

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.