# UALink **Domain:** Accelerator Interconnect / Open Standards **Doc Type:** Technical Standard Node **Maturity:** Emerging ## Definition **UALink**, Ultra Accelerator Link, is an open scale-up interconnect architecture for low-latency, high-bandwidth communication among accelerators and switches inside AI computing pods. ## Governing Boundary UALink standardizes accelerator-to-accelerator communication and associated system management rather than ordinary data-center scale-out traffic. Its load, store and atomic semantics are intended to let many accelerators participate in a larger coordinated compute and memory domain. ## Higher-Order Capability The standard changes the unit of computation. A collection of heterogeneous accelerator components can be designed around a shared scale-up fabric rather than one vendor's proprietary internal link. The organizational gain is an open boundary around the accelerator pod. ## Relation to Governance Interconnect standards allocate architectural power. They decide which devices can become peers, how the fabric is managed and where vendor differentiation remains possible. In [[wiki/Rack-Scale Computing|Rack-Scale Computing]], that decision shapes locality, fault domains and workload placement. ## Failure Modes Specification availability does not guarantee mature multi-vendor products or operational portability. Management, topology, software support and conformance are as important as link semantics. ## Related Ontology [[wiki/Standards as Governance|Standards as Governance]] · [[wiki/Rack-Scale Computing|Rack-Scale Computing]] · [[wiki/CXL|CXL]] · [[wiki/Ultra Ethernet|Ultra Ethernet]] · [[wiki/Optical Fabric|Optical Fabric]] ## Sources / Provenance - [UALink Consortium — About UALink](https://ualinkconsortium.org/about-ualink/) - [UALink Consortium — Specifications](https://ualinkconsortium.org/specification/)