The platform, called Token Fabric, was unveiled on Thursday as the company seeks to capitalise on growing demand from cloud providers and so-called neoclouds building large-scale AI data centres.
Token Fabric combines networking hardware and software to connect AI processors across a data centre, potentially reducing the need for customers to deploy and integrate separate networking systems for different chip platforms.
The company said the system is designed to improve the efficiency of expensive AI processors by ensuring they spend less time waiting for data to move through the network.
Upscale AI is targeting customers that want greater flexibility in choosing AI chips rather than relying on a single supplier as demand for computing power continues to rise.
Chief Executive Officer Barun Kar said the company expects Token Fabric to generate revenue in the tens of millions of dollars next year, with the business potentially growing into the low hundreds of millions of dollars.
The platform combines Upscale AI's own networking chips with equipment based on Nvidia's Spectrum-X technology. Its software is designed to monitor data-centre performance, identify bottlenecks and alert operators to equipment that may need to be replaced, according to the company.
The broader goal is to ensure that AI processors remain productive instead of sitting idle while waiting for data to arrive.
That issue has become increasingly important as companies spend billions of dollars building AI infrastructure and acquiring high-performance processors. Even short periods of underutilisation can reduce the return on investments in expensive AI chips and data-centre capacity.
Upscale plans to begin releasing Token Fabric in stages. The first component, which is designed to connect groups of AI chips across a data centre, is expected in the fourth quarter. The company plans to roll out the full platform incrementally through 2027.
The launch comes as cloud providers and specialised AI infrastructure companies increasingly look for ways to accommodate processors from different manufacturers.
The approach could give operators greater flexibility as the AI chip market expands beyond a single dominant supplier and as customers seek alternatives based on cost, performance, availability and specific workloads.
Upscale AI has attracted backing from major technology and investment names, including Nvidia, Salesforce and Singapore state investor Temasek.
In June, the startup raised an additional $190 million in an extension of its early-stage funding round, taking its total funding to about $500 million.
The financing was led by Premji Invest, the investment arm of Indian billionaire Azim Premji.
The latest funding and the company's new networking platform underscore the growing investor interest in the infrastructure underpinning the AI boom, as companies compete not only to develop AI models but also to build the data centres, chips, networks and software needed to run them at scale.
For Upscale AI, Token Fabric represents a push beyond individual networking components towards a broader infrastructure platform that could become a key part of how operators build and manage increasingly complex AI data centres.