The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction

What changed

Edge0 is a streaming MoE inference engine using a prerouter and recovery LoRA to serve 35B models from SSD on consumer hardware.

Why you should care

Demonstrates that routing prediction can overcome memory wall bottlenecks for large models on consumer hardware.

Your move

Watch. Monitor replication and adoption of prerouting techniques for edge inference.

Evidence

Preliminaryarxiv.org ↗

What to watch next

Independent benchmarks confirming Edge0 performance across various consumer GPU and SSD configurations.

Event

research

Event date

Sep 17, 2026

Relevant to

General AI readers