Upscale AI Launches Multi-Chip Data-Centre Fabric
Upscale AI has launched Token Fabric to connect AI accelerators from different vendors inside the same data centre.
AI data centres are becoming more heterogeneous as operators deploy accelerators from multiple vendors.
What happened
Upscale AI launched Token Fabric, a hardware-and-software networking platform designed to connect different types of AI accelerators inside the same data centre.
The goal is to let operators combine chips from multiple suppliers without building entirely separate networking environments around each one.
Why it matters
AI performance increasingly depends on how efficiently accelerators communicate, not only on the speed of an individual chip.
That makes networking and interconnect technology one of the most important bottlenecks in large AI clusters. As hyperscalers add custom chips and startups introduce specialised accelerators, interoperability becomes harder.
A neutral fabric that works across vendors could reduce fragmentation and give data-centre operators more flexibility in how they build compute infrastructure.
The bigger picture
The AI hardware stack is broadening beyond GPUs. Memory, networking, power, cooling and software orchestration are all becoming strategic layers.
Upscale AI is targeting the connective tissue between accelerators. If multi-vendor AI clusters become common, the infrastructure that lets those chips work together efficiently could become as important as the processors themselves.