Digital Twins Enable the Next Era of AI Infrastructure

Artificial intelligence (AI) is reshaping the data center. As AI workloads scale in size and complexity, traditional hyperscale designs are giving way to AI factories—purpose-built environments engineered to manufacture intelligence efficiently, reliably, and at scale.

In an AI factory, infrastructure performance is no longer measured solely by availability or power efficiency. Instead, success is defined by tokens generated per watt, workload throughput, and the ability to rapidly deploy and operate next-generation accelerated computing platforms. Meeting these demands requires a new approach to infrastructure design and operations—one grounded in system-level understanding, workload awareness, and continuous optimization.

This is where digital twins play a foundational role, and where Cadence is working closely with NVIDIA and industry leaders to enable the next generation of AI infrastructure.

From Hyperscale Data Centers to AI Factories

AI factories represent an evolution of the hyperscale model. While hyperscale data centers are designed to support a broad mix of workloads, AI factories are optimized specifically for AI training with an increasing emphasis on inference environments. These workloads introduce high rack-power densities of 100KW+ per rack, higher thermal loads, extreme AI workload variations, and critical sensitivity to network topology and latency in order to deliver the best performance per watt.

As a result, infrastructure decisions can no longer be made in isolation. Power delivery, cooling architecture (such as liquid-cooling and hybrid-cooling), workload placement, and network design are deeply interconnected. Campus-level choices—such as building layout, cooling plant strategy, and fiber routing—directly affect AI performance, efficiency, and scalability.

Designing AI factories, therefore, requires holistic, workload-specific optimization, validated before deployment, and continuously refined during operation.

Cadence and NVIDIA: Enabling AI Factory Design at Scale

As part of the effort to address these challenges, Cadence has collaborated with NVIDIA to deliver a Cadence Reality DC Elements model of NVIDIA's latest high-performance accelerated computing platform, the NVIDIA GB300 NVL72 system, for the Cadence Reality Digital Twin Platform.

This Cadence Reality DC Elements model is integrated with the NVIDIA Omniverse DSX Blueprint for AI factory digital twins and is available as a SimReady model for use via NVIDIA Omniverse libraries through the Cadence Reality DT Experience. SimReady assets provide use-case-specific technical payloads, from high-fidelity visualization and BOM data to lightweight behavioral models for rapid simulation. It enables data center designers and operators to accurately model, simulate, and optimize AI factory designs and changes during operations as part of performance-aware lifecycle management—reducing uncertainty, accelerating design cycles, and enabling confident decision-making.

By providing validated, high-fidelity infrastructure models, Cadence helps organizations design AI factories that are ready for today's workloads while remaining adaptable to future generations of accelerated computing.

Digital Twins Across the Infrastructure Lifecycle

Leading AI factory designers and operators rely on the Cadence Reality Digital Twin Platform to support both design and operational optimization. Unlike static planning tools, digital twins create a continuous feedback loop between design intent and real-world operation.

Key capabilities include:

• Operational digital twins delivered by the Cadence Reality Digital Twin Platform. Visualization and cross-team collaboration of operational digital twins is enhanced by Cadence Reality DT Experience, powered by NVIDIA Omniverse.
• Cadence Reality DC Elements models now integrated in the NVIDIA Omniverse DSX Blueprint.
• AI surrogate models that accelerate simulation and optimization, allowing teams to explore more scenarios and tradeoffs much faster, typically in minutes.
• High‑fidelity AI server and CDU models that speed end‑to‑end system design, shortening the time to deployment while optimizing AI factory architectures.

Together, these capabilities enable engineers to move beyond conservative margins and instead operate infrastructure at validated optimal points, harmonizing performance, energy efficiency, and reliability to discover new revenue opportunities.

Optimizing Infrastructure for AI Performance

In AI factories, performance is measured by the number of tokens generated per second and efficiency by tokens per watt. The Cadence Reality Digital Twin Platform enables designers and operators to optimize infrastructure directly against these metrics by simulating real AI workload behavior across power, cooling, and networking domains.

Using AI-accelerated simulations, teams can:

• Maximize token throughput by operating hardware at the most efficient tokens-per-watt point. Digital twin analysis validates running more GPUs at lower power (MaxQ), increasing token generation by up to 30% while improving overall energy efficiency (Q) and improving tokens per watt by 17%.
• Minimize cooling energy to unlock more power for compute. Optimized liquid cooling strategies, airflow, 45C inlet cooling temperatures, and thermal setpoints reduce cooling overhead and free additional power capacity for NVIDIA AI infrastructure.

By enabling high-fidelity simulation of AI workload scenarios, these capabilities accelerate gigawatt-scale AI factory buildouts and help unlock billions of dollars in potential revenue by maximizing performance while controlling energy and infrastructure costs. These results highlight the importance of treating infrastructure as a single, integrated system, validated in a high-fidelity digital twin before deployment.

Momentum Across the AI Ecosystem

Digital twins are rapidly becoming foundational to AI factory design and operations across the ecosystem:

• NVIDIA is collaborating with Cadence on Reality DC Element models for next-generation Vera Rubin technology.
• NV5 is relying on the Cadence Reality DC Design software to create digital twins that future-proof AI data centers, optimize infrastructure efficiency, and reliability. These workflows have been applied at scale across environments powered by thousands of NVIDIA Grace Blackwell GPUs, such as NVIDIA GB200, where engineering-grade, CFD-driven simulation is critical to identifying and minimizing operational risks before deployment.

Together, these examples reflect a growing industry consensus: Digital twins are essential infrastructure for AI factories.

Engineering the Future of AI Infrastructure

As AI continues to scale, the challenge is no longer simply delivering more compute—it is to efficiently convert power into intelligence. AI factories demand a new engineering mindset in which infrastructure is designed, optimized, and operated as an integrated system. With the Cadence Reality Digital Twin Platform, now integrated into the NVIDIA Omniverse DSX Blueprint, including DSX SimReady assets and AI-accelerated simulation capabilities, Cadence is helping customers design and operate AI factories with greater efficiency, speed, and confidence. The future of AI will be shaped not only by algorithms and silicon, but by infrastructure engineered to manufacture intelligence at scale.

Learn more about Cadence's partnership with NVIDIA and the Cadence Reality Digital Twin Platform.