NextFin News - South Korean semiconductor startup XCENA has secured $135 million in a Series B funding round, valuing the company at approximately $570 million as it attempts to dismantle the "memory wall" currently stifling artificial intelligence performance. The capital injection, finalized on May 29, 2026, was oversubscribed by investors betting that the next phase of AI infrastructure will be defined not by raw processing power, but by how efficiently data moves between storage and silicon. The round follows a rapid ascent for the Seoul-based firm, which rebranded from MetisX late last year and has seen its valuation nearly triple since its Series A raise in mid-2024.
The funding arrives as the semiconductor industry grapples with a widening disparity between the speed of GPUs and the bandwidth of traditional memory. While Nvidia and its peers have scaled compute capabilities exponentially, the physical bottleneck of moving massive datasets into those processors has created a performance ceiling. XCENA’s flagship product, the MX1 computational memory chip, utilizes the Compute Express Link (CXL) 3.2 standard to process data directly at the memory level. By embedding thousands of RISC-V cores into the memory controller itself, the company claims it can offload data-intensive tasks like vector database operations and big data analytics from the central processor, significantly reducing latency and power consumption.
Jin Kim, the CEO and co-founder of XCENA, has emerged as a central figure in this architectural shift. A former executive at SK Hynix and Samsung Electronics, Kim was notably the youngest vice president in SK Hynix’s history before departing to found the startup in 2022. Kim has consistently maintained that the industry’s fixation on HBM (High Bandwidth Memory) is a temporary fix rather than a structural solution. According to Kim, the "memory wall" is the primary inhibitor of AI scaling, and his long-standing position is that memory must evolve from a passive storage bin into an active participant in computation. This stance, while gaining traction among specialized infrastructure investors, remains a departure from the current GPU-centric dominance of the market.
The success of this $135 million round is partly bolstered by the South Korean government’s "K-Nvidia" initiative, a strategic program designed to foster domestic champions in the AI chip sector. This state-backed tailwind has directed significant institutional capital toward firms like XCENA, though the company’s reliance on the CXL standard places it in a competitive field. While CXL is widely viewed as the next-generation interconnect standard, it is not yet the universal consensus for all AI workloads. Some industry analysts suggest that for smaller-scale inference tasks, traditional memory architectures may remain more cost-effective, and the widespread adoption of computational memory will depend heavily on the maturity of software ecosystems that can support such decentralized processing.
XCENA has already begun shipping samples of its MX1 chip to global technology firms for proof-of-concept testing, with a more advanced MX1S model slated for 2026. The company’s roadmap includes a software development kit designed to integrate with standard data center environments, an essential step if it hopes to displace established hardware. However, the transition to a data-centric architecture faces significant hurdles, including the entrenched dominance of existing server designs and the high capital expenditure required for data centers to overhaul their current racks. Whether XCENA can translate its technical lead into a market-standard remains a question of execution rather than just engineering.
Explore more exclusive insights at nextfin.ai.
