XCENA raises $135M to attack AI inference from the memory side

XCENA adds a valuable memory-centric chip story to the feed. The Korea- and U.S.-based startup raised $135M in a Series B at a $570M valuation, bringing total funding to $185M. Its MX1 chip connects to CPUs through Compute Express Link and handles operations closer to DRAM, reducing expensive data movement between CPUs, GPUs, and memory. CEO Jin Kim argues that AI inference is increasingly a memory-scaling problem rather than only a compute problem, especially around preprocessing, KV-cache management, and data caching. The company claims workloads that used to require 10 servers could potentially run on one, though that needs real customer proof. Watch whether hyperscalers or memory vendors validate the architecture and whether CXL-attached acceleration becomes a standard way to reduce inference cost.

Key details: XCENA, $135M Series B, $570M valuation, $185M total funding, MX1 chip, CXL, DRAM, KV cache.

Continue swiping for more AI Brief stories.