Nvidia-Backed Upscale AI Launches Token Fabric Networking Tool

Upscale AI launched Token Fabric on Thursday, a hardware-software platform designed to let data centers connect AI chips from rival suppliers, with initial rollout set for the fourth quarter. The release arrives as operators seek hardware interoperability to manage supply constraints and integrate specialized accelerators alongside dominant graphics processing units.

"We expect revenue next year for Token Fabric in the tens of millions of dollars, potentially reaching the low hundreds of millions," said Barun Kar, Chief Executive Officer, Upscale AI, in a statement regarding the launch.

Platform Architecture and Capabilities

Token Fabric combines Upscale AI networking chips, equipment utilizing Nvidia's Spectrum-X technology, and diagnostic software. The integrated system identifies network bottlenecks and flags faulty equipment that may require replacement before performance degrades.

Upscale plans to release the first part of Token Fabric in the fourth quarter, with the full platform rolling out in stages through 2027. This phased release allows operators to adopt connectivity modules incrementally as they integrate non-Nvidia processors into existing infrastructure.

Funding History and Valuation

The product launch follows a $190 million funding round extension in June that valued Upscale AI at $2 billion. The extension brought the startup's total capital raised to $500 million in a financing led by Premji Invest, with additional backing from corporate and institutional investors including Nvidia, Salesforce, and Temasek.

In broader market activity, NVIDIA Corp (NVDA) stock traded at $228.52, up 0.97% as of 2026-10-13T20:00:00.000Z, according to Finnhub market data.

What Interoperability Means for Data Centers

The networking platform targets a persistent operational friction in modern artificial intelligence infrastructure, which involves combining proprietary interconnects from different silicon vendors into a unified fabric. As hyperscale operators diversify their hardware procurement across multiple semiconductor suppliers to mitigate supply chain bottlenecks, cross-vendor networking tools attempt to reduce communication latency between heterogeneous compute clusters.