Upscale AI has launched "Token Fabric," a networking platform that lets customers build data centers using AI chips from rival suppliers, according to Reuters reporting on October 8, 2026. The move positions the company at the center of one of the AI industry's most pressing problems: how to scale compute without being locked into a single chip vendor.
Upscale AI was valued at $2 billion in June and counts Nvidia, Salesforce, and Temasek among its backers. Its CEO is Barun Kar. The company expects Token Fabric to generate tens of millions of dollars in revenue next year, with potential rising into the low hundreds of millions. The first release is scheduled for the fourth quarter of 2026, with a full rollout continuing through 2027.
Why multi-vendor networking matters
The AI infrastructure market has a concentration problem. A handful of chip suppliers dominate accelerator sales, and their networking stacks are typically optimized — sometimes exclusively — for their own silicon. For data center operators and large enterprises, that creates lock-in: switching chip vendors can mean rebuilding the networking layer too. A vendor-neutral fabric that works across rival chips changes the economics, giving buyers leverage in procurement and a hedge against supply constraints from any single supplier.
This is also a timing play. Data center buildouts for AI training and inference are accelerating worldwide, and operators are making decade-long capital decisions right now. Whoever defines the networking layer for heterogeneous AI clusters stands to capture a durable, high-margin position — networking software and interconnects are among the stickiest components in the data center stack.
The investor lineup tells a story
Upscale AI's backer list is itself a signal. Nvidia's presence is the eye-catching one: the dominant AI chip supplier backing a company whose product reduces customer dependence on any single chip vendor. That apparent tension resolves if you view Token Fabric as expanding the total market — a neutral fabric lowers the barrier to building large clusters, which grows overall chip demand. Salesforce brings enterprise distribution and a cloud-scale customer's perspective on infrastructure needs, while Temasek adds sovereign-scale capital with a long horizon, the kind of backing infrastructure plays require.
A $2 billion valuation in June sets a high bar, and the company's own revenue guidance — tens of millions next year, scaling toward the low hundreds of millions — sketches the growth trajectory it needs to justify it. The Q4 2026 first release followed by a full 2027 rollout gives the company roughly a year to convert design wins into the kind of contracted revenue that supports infrastructure valuations.
Why it matters for founders and investors
For founders, Token Fabric's launch is a case study in finding the layer where value accrues. The AI boom's most obvious winners sell chips or models; the subtler winners sell the connective tissue — networking, orchestration, and tooling that makes heterogeneous infrastructure actually work. If you're building in AI infrastructure, the lesson is to look for the lock-in points that frustrate buyers and build the neutral alternative. Enterprise buyers are actively hunting for ways to diversify suppliers, and products that give them that optionality tend to get a warm reception.
For investors, Upscale AI represents the infrastructure phase of the AI cycle: after the model breakthroughs and the chip shortage, capital is flowing into the systems that make large-scale AI deployment practical and economical. The risk factors are execution — networking at data-center scale is unforgiving — and timing, since the 2027 rollout window coincides with heavy competitive investment in the same layer. But the strategic logic is sound: in a market defined by supplier concentration, neutrality is a product.
Angeled.com connects founders with investors. If you're building AI infrastructure — or any deep-tech company that needs patient, knowledgeable capital — it's worth finding investors who understand the stack.
Read Next
OpenAI Releases Codex Security as Open-Source CLI and SDK
Databricks Unveils Always-On Genie, an AI Agent That Stops Waiting to Be Asked
DeepMind Launches AlphaGenome 2, a DNA Model That Reads Genomes Cell by Cell
ElevenLabs Unveils 13 New Products and Upgrades at Its New York Summit
Google Launches Gemini Decision 1, a Model That Answers Instead of Writing
Runway Open-Sources Praxis-1, Its Robot Model That Learned by Watching Video



