Decentralized AI Model Sharing Protocol ModelMesh Approved by IETF: Devices Share AI Models Directly Without Central Servers

Decentralized AI Model Sharing Protocol ModelMesh Approved by IETF

On October 10, 2030, the Internet Engineering Task Force (IETF) officially published the ModelMesh protocol standard (RFC 9512). ModelMesh defines standards for peer-to-peer distribution and collaborative inference of AI models between edge devices, allowing smartphones, IoT devices, and vehicle systems to obtain AI models directly from nearby devices without going through centralized cloud servers.

ModelMesh's design motivation stems from a real-world problem: as AI models continue to grow in size, deploying them entirely in the cloud incurs high latency and bandwidth costs, while deploying them entirely on edge devices is constrained by storage and computing power. ModelMesh's solution is to let edge devices form an "AI model sharing network" — a model owned by Device A can be called by Device B, and Device B's inference results can be used by Device C.

The protocol's core innovation is the "model sharding" mechanism: large AI models are automatically split into multiple shards distributed across different devices. When a device needs to run a complete model, ModelMesh automatically collects shards from multiple neighboring devices and assembles them locally for execution.

Sasu Tarkoma, chair of the IETF ModelMesh working group and professor at Aalto University in Finland, said: "ModelMesh makes AI model distribution as decentralized as BitTorrent file sharing. This not only reduces cloud service costs but also enhances privacy protection — data can complete AI inference without leaving the local network."

The first platforms to support ModelMesh include Qualcomm's Snapdragon mobile platform and MediaTek's Dimensity chipset. Qualcomm stated that its flagship Snapdragon chips releasing in the second half of 2031 will include built-in ModelMesh hardware accelerators.