Token Fabric unifies scale-up, scale-out, and Upscale's AI-native software into one open, full-stack foundation for heterogeneous AI compute

• One cohesive network. Upscale Token Fabric™ connects any accelerator, scales it up inside the pod, and scales it out across the data center: one architecture, one software substrate, operated as one.
• Silicon, Systems, Software & Services. Scale-out systems powered by NVIDIA Spectrum-X; Upscale SkyFabriX™ scale-up silicon delivering 115.2 Tbps, with a roadmap to multi-petabits per second (Pb/s), powering scale-up systems supporting open standards, including OCP ESUN and UALoE. Unified by Upscale SkyOS™, one AI-native software foundation, with SkyCMD™ providing a single orchestration plane.
• Open standards end to end. Any GPU, any XPU, any NIC.

SANTA CLARA, Calif.: Upscale, Inc., the pure-play AI networking infrastructure company, today introduced Upscale Token Fabric for AI factories. Upscale’s open, purpose-built architecture unifies scale-up and scale-out networking through common software and operations, accelerating deployment with XPU partners and enabling heterogeneous AI infrastructure. Designed for hyperscalers, XPU partners, neoclouds, and enterprises, the architecture is gaining traction across customer wins and active engagements.

AI infrastructure is shifting to AI factories in which compute, networking, storage, and software must operate as one system. Yet networking vendors today offer either discrete components that customers must integrate, or systems assembled from legacy networking components not purpose-built for AI. As heterogeneous XPUs, rack-scale architectures, and inference workloads drive unprecedented demands, the market needs a purpose-built, full-stack, open and standards-based AI networking solution. It must unify scale-up, scale-out, and orchestration to improve utilization, simplify operations, and optimize tokens per dollar.

The opportunity is massive: Dell'Oro forecasts nearly $1 trillion in AI back-end switch spending through 2030, with Ethernet becoming the leading scale-up technology. 650 Group projects AI networking will surpass $200 billion by 2030 as heterogeneous, full-stack solutions scale.

Upscale Token Fabric is an industry-first open architecture designed to:

• Connect any accelerator
• Scale accelerators inside the pod
• Scale accelerators across the data center

Together, these capabilities provide the networking foundation that enables AI factories.

Upscale Token Fabric brings together:

• Scale-up: SkyFabriX silicon and switch trays
• Scale-out: Upscale-engineered systems powered by NVIDIA Spectrum-X
• Operate as one: SkyOS and SkyCMD unify software and operations across GPUs and heterogeneous XPUs

“The industry has spent decades connecting computers. AI demands something different: one computer, built at data-center scale. Upscale Token Fabric brings heterogeneous accelerators, scale-up, scale-out and software together as one coherent AI system, built on open standards, so the network never gets in the way. Operators focus on delivering the best AI compute for their workloads; the integrated stack delivers the performance and time to market,” said Barun Kar, CEO, Upscale.

Upscale Token Fabric: Building blocks for AI factories

SCALE UP

• SkyFabriX: Industry’s first purpose-built scale-up switch silicon based on previously announced SkyHammer™ architecture, supporting ESUN, UALoE, SUE-T, standard Ethernet/IP, and evolving protocols including UALink, connecting any GPU or XPU over one open fabric. Engineered for deterministic, ultra-low-latency and native memory-semantic communication with a multi-Pb/s roadmap, including copper and optical variants (pluggable, NPO, CPO).
• Scale-up switch-tray systems: Custom, ORV3 and standard rack form factors that pool GPUs and heterogeneous XPUs into large, standards-based domains.

SCALE OUT

• Scale-out switch systems: NVIDIA Spectrum-X-powered scale-out systems, 400G and 800G to 1.6T, ready for future Spectrum-X generations. With SkyOS, standard and custom system form-factor variants integrate seamlessly into NVIDIA and heterogeneous XPU clusters.

OPERATE AS ONE

• SkyOS: An AI-native network operating system that provides one software foundation across scale-up and scale-out fabrics. Built from the ground up for AI factories, it delivers unrivaled advantages in performance, resiliency, and analytics.
• SkyCMD: Agent-native orchestration and observability that enables thousands of network elements to observe, reason, and act as one, creating a unified operational model across scale-up and scale-out fabrics.

Consume It Your Way

Customers can adopt the solution through flexible consumption models. For scale-up, Upscale systems are built on Upscale’s own SkyFabriX silicon. For scale-out, Upscale-engineered systems are built on NVIDIA Spectrum-X silicon. Customers can choose to adopt silicon, systems, software, or the full stack.

• Full stack-maximum simplicity: Standard form factors for scale-up switch trays and scale-out systems with SkyOS and SkyCMD managing the complete rack: one full-stack operating model, from deployment through day-two operations. Built for neoclouds and enterprises without hyperscaler-size network teams.
• Integrated systems-flexible integration: Standard form factors for scale-up switch trays and scale-out systems, with SkyOS and SkyCMD optionally managing the complete rack. Built for XPU partners that want validated hardware without building a network stack.
• Silicon, systems, your software-maximum control: SkyFabriX scale-up silicon, custom scale-up switch trays, or custom scale-out switch systems with a compatible SDK and open SAI from Upscale, and the customer's own NOS integrated on top. Built for hyperscalers that design their own systems and operations.

Ecosystem Support

“The network is the backbone of the AI factory, keeping accelerators connected and productive as systems scale,” said Gilad Shainer, senior vice president of networking at NVIDIA. “By using NVIDIA Spectrum-X Ethernet for Token Fabric’s scale-out systems, Upscale is helping customers build high-performance, open and heterogeneous AI infrastructure.”

“AI infrastructure is driving unprecedented demand for high-performance, energy-efficient networking silicon. We look forward to continuing our collaboration with innovators like Upscale, leveraging TSMC’s leading-edge process and advanced packaging technologies as they deliver its innovative AI networking infrastructure solutions to the market.” -Lucas Tsai, Vice President of New Strategic Engagements at TSMC North America

“Inference at scale demands open, standards-based fabrics that let specialized XPUs and compute platforms do their best work. An open scale-up standards architecture with ESUN support like SkyFabriX enables customers to easily incorporate d-Matrix’s memory-centric XPUs and rack-scale solutions into their data centers for applications that require higher levels of interactivity.” -Sid Sheth, Founder & CEO, d-Matrix

“As AI models grow larger and agentic workloads become more demanding, scaling tightly coupled accelerator systems requires higher-bandwidth, lower-latency, and more efficient scale-up networking. Meeting that demand will require architectures purpose-built for the AI era.” -Rodrigo Liang, Co-Founder & CEO, SambaNova

“As AI infrastructure becomes more complex and expands beyond the hyperscalers, there’s a growing need for integrated, full-stack architectures that deliver the required performance and time to market,” said Sameh Boujelbene, Vice President at Dell’Oro Group. “Upscale Token Fabric is addressing this demand, bringing scale-up and scale-out networking together in a unified architecture for heterogeneous AI infrastructure.”

“As AI workloads scale, operators need greater bandwidth and lower latency to keep accelerators productive. Upscale Token Fabric brings together purpose-built silicon, systems and software to address the performance and operational needs of heterogeneous AI infrastructure.” -Alan Weckel, Founder and Technology Analyst, 650 Group

Availability

General availability is planned for early 2027, with early-access and joint-validation programs underway today. Customers and ecosystem partners can learn more at upscale.com and get a first look at Upscale’s Token Fabric and systems at OCP 2026, booth Sapphire F5.

About Upscale

Upscale is a pure-play AI networking infrastructure company delivering an open, high-performance, standards-based networking portfolio for heterogeneous AI compute. Its purpose-built, integrated portfolio spans AI scale-up, scale-out and AI-native software and orchestration, enabling hyperscalers, XPU providers, neoclouds, and enterprises to optimally deploy and operate AI infrastructure as one cohesive system. For more information, visit upscale.com.

Forward-Looking Statements

This press release contains forward-looking statements regarding Upscale’s products, roadmap, performance targets, and availability. These statements reflect current expectations and are subject to risks and uncertainties, including changes in development timelines, product performance, manufacturing delays, supply chain constraints, and market conditions. Actual results or performance may differ materially from those expressed or implied. Forward-looking statements speak only as of the date of this release, and Upscale undertakes no obligation to update them except as required by law.

Fonte: Business Wire