Majestic Labs bets on LPDDR6 to break AI memory bottlenecks

Majestic Labs, a Tel Aviv-based startup founded in 2023 by former Google and Meta engineers, has unveiled Prometheus, a server designed to replace Nvidia-style GPU and HBM systems for AI inference. The system uses Ignite AI Processing Units that combine Arm cores with RISC-V vector and tensor engines, aiming to address workloads the company says are increasingly constrained by memory capacity.

Each Prometheus server can house up to 12 AIUs and share between 8 TB and 128 TB of LPDDR6 memory as one contiguous, coherent pool. Custom memory aggregation chiplets connect that memory over copper cables up to one metre long, while a standard 40U rack holds four servers, draws 120 kW and uses cold-plate liquid cooling. Majestic compares the design with an Nvidia DGX B300 system offering 2.3 TB of HBM3e plus up to 4 TB of DDR5 system memory, claiming over 50 times more fast memory and 1.7 times the interconnect bandwidth.

The company says Prometheus could cost between 10 and 50 times less than a GPU system of equivalent performance when it ships next year, with support for PyTorch, vLLM and OpenAI’s Triton. Majestic employs around 40 people across Tel Aviv and Los Angeles and raised $100 million in an A-round late in 2025, but its performance claims still await independent benchmarking and shipped hardware.