# Fireworks AI explains splitting prefill and decode across datacenters and hardware

> Original title: Matt Murphy (@FireworksAI_HQ) on separating prefill and decode across datacenters and hardware classes.

- Source: LMSYS Org
- Published: 2026-11-09T23:15:41.000Z
- Bad Influence: https://badinfluence.news/items/hygj25zj2hsxjj7wdf8qmv8q2
- Original: https://x.com/lmsysorg/status/2119931966640525686

## Summary

Matt Murphy of Fireworks AI describes separating prefill and decode across datacenters and hardware classes. He names the trade-offs of moving KV cache over limited Ethernet links, reworking request lifecycles, and securing traffic across datacenter boundaries.

## Post

Matt Murphy (@FireworksAI_HQ) on separating prefill and decode across datacenters and hardware classes.

The trade-offs in moving KV cache over limited Ethernet links, reworking request lifecycles and securing traffic across datacenter boundaries.

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHWuCNcjbYAABcuJ.jpg&mode=thumb&exp=1794528000&sig=db583b613f203327)