XCENA Raises $135M — AI’s Bottleneck Is Memory, Not Compute
XCENA raised $135M Series B at $570M valuation for MX1, a chip placing compute inside DRAM to eliminate the memory round trip bottlenecking AI inference.
In Brief
- XCENA raised $135M Series B at a $570M valuation to build the MX1, a memory chip that places compute directly inside DRAM modules.
- Founded by Samsung and SK Hynix veterans, the startup argues AI inference is bottlenecked by memory bandwidth — not GPU supply.
- The MX1 prototype packs thousands of RISC-V cores alongside 2TB of DRAM, with mass production on Samsung’s 4nm line set for late 2026.
Every time ChatGPT generates a word, data shuttles between memory, a CPU, and a GPU — then back again. That round trip repeats for every single token. TechCrunch reports that XCENA, a startup with offices in South Korea and Sunnyvale, just raised $135 million to eliminate that shuttle.
The Series B values XCENA at $570 million and brings total funding to $185 million. Seoul-based VC firms Atinum Investment and IMM Investment co-led the round, with Corstone Asia, SBI Investment, and Mirae Asset Capital also participating, per SiliconANGLE.
XCENA’s chip, the MX1, connects to the CPU via CXL (Compute Express Link) and handles data orchestration, KV cache management, and preprocessing directly inside the memory module — before data ever reaches the GPU. The company claims what previously required 10 servers could run on one.
Memory-First Architecture
CEO Jin Kim co-founded XCENA in 2022 with CTO Dohun Kim and CPO Harry Juhyun Kim, all veterans of Samsung and SK Hynix. Kim told TechCrunch that CPUs and GPUs improved over decades while memory did not — and that inference is increasingly a memory scaling problem.
The MX1 combines up to 2TB of DRAM with thousands of RISC-V cores in four-core clusters, each with dedicated L1 cache. It handles KV cache refreshes, data caching, and preprocessing — freeing GPU cycles for the matrix math they excel at.
XCENA targets hyperscalers spending tens of billions annually on AI infrastructure. Samsung, SK Hynix, and Micron each crossed a $1 trillion valuation this month. Samsung’s own profit surge was driven by AI memory demand, as Frontierbeat reported.
Rivals include Nasdaq-listed Astera Labs and Marvell. Kim said XCENA’s thousands of purpose-built cores differentiate it from Marvell’s handful of general-purpose ones.
FAQ
What is XCENA’s MX1 chip?
The MX1 is a computational memory controller that places thousands of RISC-V cores inside DRAM modules, handling data orchestration and KV cache management without routing through CPUs or GPUs.
How much did XCENA raise?
XCENA raised $135 million in a Series B round at a $570 million valuation, bringing total funding to $185 million.
When will the MX1 enter production?
Mass production on Samsung’s 4nm process is scheduled for late 2026, with revenue expected starting in 2027.
Who are XCENA’s competitors?
Astera Labs and Marvell, both Nasdaq-listed companies working on next-generation memory connectivity, are the closest rivals.
XCENA expects to generate revenue starting in 2027.
