Accelerating the AI Factory: Switch and Cadence Redefine High-Density Design

"We are redefining what is possible for next-gen AI factories with our patent-pending EVO Chamber solution—delivering up to 2MW per cabinet through advanced hybrid cooling in a modular, future-proof design. Using the Cadence Reality Digital Twin Platform, built with NVIDIA Omniverse libraries, Switch is developing a 5‑star DC Element—a physics‑accurate behavioral model and foundational building block for AI factory digital twins—of its EVO Chamber. This EVO Chamber Cadence Reality DC Elements model enables seamless design integration and rapid qualification of IT systems, including NVIDIA GB200 and GB300 NVL72 platforms, across a wide range of operating scenarios. Validation is performed using coupled physics‑based computational fluid dynamics (CFD) and flow network modeling (FNM) co‑simulation, allowing designs to be verified well before a single cabinet is installed."
- Zia Syed, Chief Technology Officer, Switch

Introduction

As AI factories evolve toward multi-megawatt rack densities, leading operators are turning to digital twins to design, validate, and maximize workload throughput before deployment. Switch is advancing this approach by using the Cadence Reality Digital Twin Platform, leveraging NVIDIA Omniverse libraries, to model AI factory environments in which IT equipment, power delivery, and cooling systems are developed and simulated together.

This digital-first methodology now extends beyond system-level AI factory design into the physical infrastructure itself—most notably the patent-pending Switch EVO Chamber^®, purpose-built to support extreme rack densities and next-generation NVIDIA GB200 NVL72 and NVIDIA GB300 NVL72 platforms. By expanding digital twin validation from the data hall to the cabinet and cooling architecture, Switch and Cadence are redefining how high-density AI infrastructure is designed, tested, and deployed.

The Density Challenge

Traditional data center designs were never built for the extreme requirements of next-generation generative AI. The sheer power density and variable power demands of modern GPU clusters that go beyond 100kW+ per rack, such as those powered by NVIDIA DGX SuperPOD, require a fundamental rethink of data center architectures, giving the rise of AI factories.

Switch's EVO Chamber is purpose-built for this new era. It enables AI factories to scale far beyond the power and cooling limits of traditional data halls, supporting the immense heat rejection requirements of liquid-to-chip and high‑performance air‑cooled systems. Through advanced hybrid cooling, each modular, future‑proof cabinet can support up to 2MW, unlocking unprecedented density while maintaining operational resilience. However, implementing such advanced infrastructure requires precise planning. Integrating these chambers into existing or greenfield sites introduces complex variables across airflow, fluid dynamics, power delivery, cabling, and spatial constraints.

Simulating the Future of Infrastructure

This collaboration empowers organizations to de-risk and maximize token throughput for their AI investments through advanced simulation. The new Cadence Reality DC Elements model of the EVO Chamber allows engineers to drag and drop them directly into any data hall layout within the Cadence Reality Digital Twin Platform.

Once placed, these digital twins can be populated with specific cabinet payloads, including NVIDIA DGX GB200 and NVIDIA DGX GB300 systems. This is not a static model, but a physics‑based simulation of a real-world counterpart that predicts token‑production efficiency. By quantifying how power is consumed across compute and infrastructure, Cadence Reality Digital Twin Platform enables optimization of tokens per watt—maximizing power used for AI workloads within a fixed power envelope. This capability transforms facility planning from a static estimation exercise into a dynamic engineering discipline. Leaders can now answer critical questions with certainty:

• How will mixed cooling topologies interact within the same hall?
• Do we have sufficient operational headroom for peak workloads?
• Are the liquid-to-chip setpoints optimized for energy efficiency?

Optimizing Cooling Architectures

One of the most significant advantages of this integration is the ability to validate complex cooling strategies in software. The Switch EVO Chamber details five separate fluidic system pathways designed to support both liquid-to-air and liquid-to-chip cooling strategies.

Aligning these pathways with modern direct-to-chip strategies for high-power GPUs is complex. Through the Cadence platform, operators can simulate and compare multiple cooling configurations side by side. This enables the optimization of performance, efficiency, and resiliency in a virtual environment, ensuring capital is deployed effectively and the physical facility is optimized from day one.

Interoperability and the Open Ecosystem

Time to first token is a competitive advantage in the AI market, and silos slow down innovation. Recognizing this, the collaboration emphasizes interoperability. Switch maintains OpenUSD interoperability across its internal design environment.

This OpenUSD readiness enables a connected asset pipeline that links layout, visualization, and simulation across different teams and tools, including NVIDIA Omniverse libraries. It bridges the gap between facilities engineering, IT operations, and executive decision-making, providing a single source of truth for the entire organization.

Conclusion: Validation Before Deployment

In the high-stakes world of AI factories, and design flow inefficiencies are liabilities. The integration of Switch's EVO Chamber into the Cadence Reality Digital Twin Platform provides a comprehensive toolkit for designing higher-performing AI factories faster. By enabling rigorous validation of placement, density, and cooling topology before deployment, Switch and Cadence are helping organizations build not just faster, but smarter.

This is more than a modeling tool; it is a strategic asset for operational efficiency. It ensures that, as we scale into the multi-megawatt future, our infrastructure is as intelligent as the AI models it supports.

Learn more about the Cadence Reality Digital Twin Platform and Switch EVO Chamber.