Investors
Inference hardware for 600B–3T models, built on flash instead of HBM.
Why now
Frontier open models are mixture-of-experts, and serving them is limited by HBM and GPU supply. Flash is plentiful, cheap per gigabyte and fast when read across many channels.
NANDmind turns that into an inference accelerator for 600B–3T-parameter models.
Stage
Pre-seed. Two US provisional patent applications filed. Trace-driven simulator in progress. Early-access list open.