A new card for the data center
Qualcomm announced its Cloud AI 100 Ultra AI accelerator card. The product sits in Qualcomm's data center lineup and is positioned for AI inference workloads.
The headline numbers
The memory side stands out: 128 GB of capacity and 548 GB/s of bandwidth. Power consumption is listed at 150W. In AI models, memory capacity and bandwidth directly determine how much of a model can run on a single card.
Why it matters
Inference costs for large language models often hit a memory bottleneck in hardware. That is the area Qualcomm is targeting: keeping more of a model in memory with less power. The company has recently been leaning on its data center business as smartphone sales fluctuate.
