Netpreme

Netpreme Corporation develops AI-native memory systems designed to connect memory and compute more efficiently, from silicon to software. Its X-Mem™ Networked Memory Tier is built around the XM-100 Memory Processing Unit (MPU), a memory expander intended to work alongside AI compute. Netpreme also describes a composable architecture in which MPU memory nodes and XPU compute nodes share capacity through a scale-up fabric.

The company’s offerings include XM-100 MPU silicon, a memory-augmented virtual machine with composable GPU memory, and XM-Serve™ inference for open models and coding agents. Its software includes an X-Mem library and memory-tiering capabilities designed to integrate with GPU kernels and inference or training runtimes. The company identifies KV and prefix caching, long-context inference, multi-model serving, and optimizer or activation offloading as supported use cases. Netpreme reports performance results including faster time to first token than DRAM offload on prefix-heavy workloads and higher concurrency using the same number of GPUs.