Majestic Labs builds disaggregated memory-compute servers — its MLS1 pairs a custom Ignite chip with ~128 TB of unified DRAM to break the AI memory wall for LLM inference, with no Nvidia GPUs and no HBM.
Roughly 5× lower cost per terabyte than HBM, dropping into legacy 30 kW data centers with no model rewrites. A team drawn from Google TPU and Meta silicon, 120 patents, a $600M+ pre-tapeout order book, and ranked the #1 most promising Israeli startup of 2026.