28 May 2026

Primemas Eliminates Memory Bottleneck to Maximize GPU Utilization

Stand: Booth 213
Primemas, a fabless semiconductor innovator headquartered in Silicon Valley, is set to begin mass production of JBOM (Just a Bunch of Memory) in the second half of 2026 ; a CXL- based solution that expands server memory capacity from the conventional 2 - 4 TB limit to hundreds of terabytes per CPU server.
Designed to tackle the memory bottleneck that prevents expensive GPUs from being fully utilized in AI data centers, JBOM delivers a scalable memory pool architecture that directly translates to higher GPU utilization and AI workload performance. 

The company has secured mass supply agreements with Micron and is co-developing next-generation CXL memory solutions with Samsung Electronics. Alibaba's Beluga research validates the approach, demonstrating a 7.35x throughput improvement with CXL-based memory expansion. 

With a roadmap to expand capacity beyond 240 TB by 2027 and $72 million raised in its 2025 Series B round,
Primemas is positioning itself at the forefront of the memory-centric computing era. 
Loading