Qualcomm Cloud AI 100 Ultra product brief

Official single-page brief, fetched and text-extracted in full on 2026-10-03. Full spec list as printed: 'Form factor: PCIe FH3/4L', 'TDP: 150W', 'ML capacity (INT8): 870 TOPs', 'On-die SRAM: 576 MB', 'On-card DRAM: 128 GB LPR4x 548 GB/s', 'Host interface: PCIe Gen 4, 16 Lanes', 'Number of cores: 64 AI cores on single card'. Marketing on the sheet: 'Best Perf/TCO$', '100B Gen AI models on a single card', '8x larger models within a single server'. Same conventions as the base brief: INT8 TOPS only, no TFLOPS, no bus width, no process node, no sparsity statement. The 548 GB/s is printed next to the 128 GB capacity in one cell, again without a bus width, so it cannot be decomposed into bus x pin rate. Note 'LPR4x' is Qualcomm's own typo/romanisation of LPDDR4X throughout.