XCENA raises $135M Series B at $570M valuation for memory-centric AI chip
What's the deal? South Korean startup Xcena has raised $135 million to develop its MX1 chip, a memory-centric processor designed to tackle what it sees as AI's real bottleneck: getting data to processors fast enough. The Series B round values the company at $570 million.
Rather than shuttling data between separate memory and processing units, the MX1 embeds thousands of RISC-V cores and vector engines directly alongside memory — an approach called near-data processing. Xcena claims the chip can deliver up to 3.9x improvements in time-to-first-token, a key metric measuring how quickly a language model starts generating a response.
The company, formerly known as MetisXDealroom has a profile for this one. Try Dealroom →, was founded in 2022 and rebranded in late 2024. It operates as a fabless semiconductor firm — designing chips but outsourcing manufacturing, like AMD, Nvidia, and Qualcomm. CEO Jin Kim leads the company, which had previously raised roughly $50 million across a seed round and Series A.
Working samples of the MX1 are targeted for October 2025, with a production version planned for 2026.
Why now? Memory bandwidth and capacity have become critical bottlenecks for large language models, which need to store and access enormous amounts of data during inference. As models get longer context windows, the memory structures they use to store conversation context — called key-value caches — balloon in size, creating a genuine engineering challenge.
This dynamic has already fuelled a boom in high-bandwidth memory, where SK HynixDealroom has a profile for this one. Try Dealroom → holds roughly 57% of the market. Xcena's approach is architecturally different: by putting processing power directly at the memory level, the MX1 aims to eliminate the data movement problem altogether.
What could go wrong? The 3.9x improvement claim is eye-catching, but benchmark numbers from startups deserve scepticism until validated in real-world deployments. SK Hynix and Samsung aren't standing still — both are investing heavily in next-generation memory and exploring their own computational memory strategies.
Xcena also faces the classic challenge of any fabless chip startup: convincing data centre operators to adopt a new architecture when incumbents already have deep ecosystem integration.
The signal: Xcena's round highlights a broadening of AI infrastructure investment beyond pure compute. While GPU makers have dominated the narrative, the memory bottleneck is drawing serious capital — and at a $570 million valuation for a company without production silicon, investors are pricing in the conviction that memory architecture will be a decisive layer in the AI stack. Dealroom already classifies Xcena as a late-growth company, a notable trajectory for a firm founded just three years ago, underscoring how quickly the market for specialised AI hardware is maturing.
Read more: cryptobriefing.com
