The brief

Analyst view

• Corsair: SRAM-based in-memory compute. 10x faster inference vs. GPUs; 80% lower power.
• June 2026: Full production; shipping to hyperscalers, neoclouds, and frontier labs.
• Pairs with NVIDIA Blackwell GPUs: GPUs do prefill, Corsair excels at decode phase.

Technical approach

As reported

Compute-in-memory (CIM)

Math inside the memory array

Computation within high-bandwidth SRAM eliminates data movement bottleneck. Delivers 10x better decode performance.

Heterogeneous inference

GPUs for prefill, Corsair for decode

Prefill is compute-heavy; decode is memory-bound and latency-critical. Corsair specializes in decode.

Memory hierarchy

SRAM for speed, LPDDR5 for capacity

~150 TB/s SRAM + up to 256GB LPDDR5. Models fit on single card or scale across rack.