XCENA raises $135M in Series B

XCENA Inc. has successfully raised $135 million in a Series B funding round aimed at advancing its computational memory technology, which is designed to enhance the performance of artificial intelligence clusters. The round was spearheaded by Korean investment funds Atinum Investment and IMM Investment, with additional backing from more than six other institutional investors. This latest influx of capital elevates XCENA's valuation to $570 million, underscoring strong investor confidence in its innovative approach to tackling bottlenecks in AI processing.

Founded in 2022 by former Samsung Electronics Co. and SK hynix Inc. employees, XCENA focuses on developing cutting-edge memory solutions for AI applications. The company's flagship product, the MX1, is a computational memory controller engineered to facilitate AI inference workloads by speeding up data management tasks. The device is designed to store an AI model's key-value (KV) cache and vector databases directly, circumventing the performance issues associated with using slower external Dynamic RAM (DRAM). This feature allows for improved performance by enabling the reuse of KV caches across multiple requests, thereby reducing processing demands.

The strategic intent behind the MX1 is to tackle the latency challenges faced by large language models (LLMs) when their required KV cache exceeds the capacity of a graphics card's memory. By integrating up to two terabytes of DRAM with thousands of central processing unit cores, the MX1 addresses this issue, promising substantial boosts in inference performance. Furthermore, the MX1 is capable of accelerating data-heavy analytics applications like Apache Spark by minimizing data travel times between computational units and memory, thanks to its architecture derived from the open-source RISC-V framework.

In a broader market context, XCENA’s technology could significantly impact the computational memory sector by providing a sophisticated solution to enhance AI and analytics workloads. This positions XCENA as a potential disruptor amidst increasing demand for efficient memory solutions within AI infrastructures, challenging established memory hardware providers. The company intends to allocate the raised capital towards developing further computational memory innovations and strengthening its market presence through partnerships, particularly with hyperscale cloud providers.

Looking forward, XCENA plans to commence mass production of the MX1 by the end of the year using Samsung's advanced four-nanometer process. Initial revenue is anticipated in 2027. As the company continues to refine its product and build industry alliances, XCENA's progress will be closely watched by stakeholders looking to capitalize on advancements in AI processing technology.