Xcena Raises $135M to Kill the AI Memory Bottleneck
Xcena's MX1 chip handles data orchestration inside memory modules, cutting out the middleman in AI inference.
Every time you fire off a prompt to ChatGPT, your data goes on a relay race — bouncing from memory to CPU and back before anything useful happens. Xcena wants to end that nonsense.
The startup just closed a $135M Series B at a $570M valuation. Its secret weapon: the MX1 chip, which performs data orchestration and KV cache management directly within memory modules instead of shuttling everything through a processor first.
KV cache is the mechanism that lets large language models remember context during a conversation. Managing it efficiently is a massive bottleneck in AI inference workloads. Xcena's approach embeds that logic right where the data already lives.
It's a clever architectural bet. If AI inference demand keeps scaling the way everyone expects, eliminating memory-to-CPU round trips could matter a lot.