Tech

Xcena secures $135 million Series B to challenge AI memory bottlenecks

Backed by $570 million valuation, the firm plans mass production via Samsung’s foundry by late 2026, aiming to reduce server requirements for AI workloads.

Author
Owen Mercer
Markets and Finance Editor
Published
Draft
Source: TechCrunch · original
This chip startup just raised $135M on a bet that AI’s biggest bottleneck isn’t compute — it’s memory
South Korean chip startup targets infrastructure inefficiencies with memory-centric MX1 architecture

South Korean semiconductor startup Xcena has closed a $135 million Series B funding round, valuing the company at $570 million. The investment brings the firm’s total capital raised to $185 million since its inception in 2022. Co-led by Seoul-based venture capital firms Altinum and IMM Investment, the round also included participation from Corstone Asia, SBI Investment, and Mirae Asset Capital.

The company is developing the MX1 chip, a memory-centric architecture designed to address latency issues inherent in current AI infrastructure. By integrating compute capabilities closer to dynamic random-access memory (DRAM), the technology aims to process data before it leaves the memory module. This approach seeks to eliminate the inefficient data relay race between CPUs, GPUs, and memory that characterises traditional AI processing workflows.

Xcena’s chief executive, Jin Kim, stated that the MX1 chip connects to the central processing unit via Compute Express Link (CXL). This dedicated interface allows the chip to handle routine data operations, such as preprocessing and key-value cache management, directly within the memory module. The company claims this efficiency could reduce the number of servers required for certain tasks from 10 to just one.

Founded by former executives from Samsung and SK Hynix, Xcena differentiates its technology through the use of thousands of small, efficient RISC-V cores. This contrasts with rivals such as Marvell and Astera Labs, which typically rely on fewer general-purpose cores. The startup also maintains vertical integration by designing its own internal memory hierarchy and DRAM controller, rather than outsourcing these components.

Mass production of prototype chips is scheduled to begin via Samsung’s foundry by the end of 2026, with revenue generation expected to commence in 2027. Xcena currently employs more than 90 staff across offices in Pangyo, South Korea, and Sunnyvale, United States. The firm is also in early-stage conversations with several global memory vendors regarding potential partnerships.

Continue reading

More from Tech

Read next: Apple to roll out manual EQ controls for AirPods in iOS 27 update
Read next: Apple rolls out visionOS 27, integrating AI-driven Siri into Vision Pro headset
Read next: Apple Overhauls Siri with Google Gemini Partnership and Standalone App at WWDC 2026