Majestic Labs launches Prometheus AI server with up to 128TB memory

Majestic Labs, a Tel Aviv‑based startup founded in 2023 by former Google and Meta engineers, introduced the Prometheus server as an alternative to Nvidia GPU‑HBM systems for AI inference. The server replaces GPUs with up to 12 Ignite AI Processing Units that combine Arm cores with RISC‑V vector and tensor engines, and it aggregates 8 TB to 128 TB of LPDDR6 memory into a single coherent pool accessed via custom chiplet interconnects.

A standard 40U rack can hold four Prometheus servers, drawing about 120 kW and using liquid‑plate cooling. Majestic claims the architecture delivers more than 50 times the fast memory of an Nvidia DGX B300 and 1.7 times higher interconnect bandwidth, while costing 10‑50 times less and consuming less electricity. The company raised $100 million in a Series A round in late 2025 and says it already has orders from large enterprises and hyperscalers.