Upscale AI Unveils Token Fabric Platform to Connect AI Chips from Multiple Suppliers

Upscale AI Launches Token Fabric to Bridge AI Chip Interoperability

Nvidia-backed Upscale AI unveiled Token Fabric on Thursday, a networking platform designed to connect AI chips from different suppliers within data centers1

1

2

2

. The platform combines hardware and software to link AI processors across facilities without requiring customers to integrate separate networking systems. Built on open standards including UALink and Ethernet for Scale-Up Networking (ESUN), Token Fabric enables chips from rival makers to operate on the same network, addressing a critical challenge as AI systems scale1

1

.

Platform to Connect Chips Tackles Data Bottlenecks

Large AI systems distribute computational work across thousands of chips that must constantly exchange data. Networking gear links chips within a single server rack, known as scale-up, and connects racks across buildings, known as scale-out. When data moves too slowly, costly chips sit idle1

1

. Token Fabric addresses these data bottlenecks by combining SkyFabriX switch chips, Upscale's own scale-up technology, with scale-out switches built on Nvidia's Spectrum-X Ethernet chips1

1

. Two software layers, SkyOS and SkyCMD, run and monitor the entire network, identifying bottlenecks and flagging hardware that may need replacement1

1

2

2

.

Revenue Projections and Market Strategy

CEO Barun Kar told Reuters he expects Token Fabric revenue next year in the tens of millions of dollars, potentially reaching the low hundreds of millions2

2

3

3

. The Santa Clara startup is targeting large cloud providers and neoclouds, smaller firms that rent out AI computing power1

1

3

3

. The aim is to keep expensive AI chips working rather than waiting for data, helping customers maximize their investment2

2

.

Deployment Timeline for AI Data Centers

Upscale plans to release the first part of Token Fabric, which connects groups of AI chips across a data center, in the fourth quarter of this year2

2

3

3

. The full platform will roll out in stages through 2027, with general availability expected in early 20271

1

. Early-access programmes are already under way1

1

.

Funding and Strategic Backing

Barun Kar and executive chairman Rajiv Khemani founded Upscale AI in 2025. The company raised a seed round of more than $100 million in September and a $200 million Series A in January1

1

. In June, it added a $190 million extension led by Premji Invest, the investment unit of Indian billionaire Azim Premji, valuing the startup at $2 billion1

1

2

2

. Nvidia, Salesforce Ventures, and Temasek joined that round, bringing total funding to approximately $500 million1

1

3

3

. Nvidia has taken stakes in many AI firms, building a $99 billion equity portfolio by late July1

1

.

Industry Implications and Competition

Gilad Shainer, Nvidia's senior vice president of networking, stated that "the network is the backbone of the AI factory, keeping accelerators connected and productive as systems scale"1

1

. The launch positions Upscale AI to tap growing demand from customers building out AI data centers and seeking to integrate AI chips from multiple suppliers2

2

. Other startups are also raising capital to ease data bottlenecks between chips. In July, chip interconnect firm Eliyan reached a $1 billion valuation1

1

. Watch for how established players respond to this push for AI chip interoperability and whether open standards gain traction over proprietary solutions like Nvidia's NVLink.