XCENA Secures $135M Series B to Bring Compute to Memory

Every time an artificial intelligence model processes a request, data travels from memory through a central processing unit, to a graphics processing unit, and back again. This continuous routing represents one of the most expensive and power-intensive structural bottlenecks in the current AI infrastructure boom.

XCENA, a four-year-old semiconductor startup with offices in Sunnyvale, California, and Seoul, South Korea, is attempting to eliminate this costly round trip. The company has secured $135 million in Series B funding at a $570 million valuation to develop chips that handle data operations directly next to the memory modules.

Jin Kim, Chief Executive Officer of XCENA, co-founded the company in 2022 alongside other veterans from Samsung and SK Hynix. The startup operates on the premise that inference processing is fundamentally becoming a memory scaling problem rather than purely a compute challenge. The recent funding brings their total capital raised to $185 million.

The company’s MX1 chip utilizes Compute Express Link technology to establish a dedicated pathway between processors and memory. By managing tasks such as data orchestration, preprocessing, and cache management directly within the memory module, the system reduces reliance on general-purpose processors for routine data operations. The company indicates this architecture could allow workloads that previously required multiple servers to run on a single unit.

Seoul-based venture firms Atinum and IMM Investment co-led the financing round. The capital will support the development and commercialization of the MX1 prototype. Mass production of the chips is scheduled to begin on Samsung foundry lines by late 2026, with the company anticipating initial revenue generation in 2027.