source: techcrunch ai: this chip startup just raised $135m on a bet that ai’s biggest bottleneck isn’t compute — it’s memory

level: technical

xcena, a four-year-old chip startup with offices in south korea and sunnyvale, raised $135 million in series b funding at a $570 million valuation. the round was co-led by altinum and imm investment, with participation from corstone asia and existing backers sbi investment and mirae asset capital. total funding now stands at $185 million. the company was founded in 2022 by veterans from samsung and sk hynix, the dominant memory chip makers.

the startup's mx1 chip uses compute express link to connect directly to cpus, processing data inside the memory module instead of shuttling it back and forth to gpus or cpus. this design targets the overhead of preprocessing, kv cache management, and data caching that typically runs on general-purpose processors. xcena claims its approach could reduce the server count needed for some workloads from ten to one. the chip contains thousands of small, efficient risc-v cores and a custom memory hierarchy, interconnect, and dram controller.

mass production is planned for late 2026 through samsung's foundry, with revenue expected in 2027. the company is in early talks with global memory vendors and aims to sell to hyperscalers spending heavily on ai infrastructure. rivals include astera labs and marvell, but xcena differentiates through its high core count and vertical integration. the funding comes as memory-centric architectures gain attention amid rising memory prices and trillion-dollar valuations for samsung, sk hynix, and micron.

why it matters: reducing data movement between memory and processors could lower the cost and energy use of ai inference, making large-scale deployment more efficient.


source: techcrunch ai: this chip startup just raised $135m on a bet that ai’s biggest bottleneck isn’t compute — it’s memory