Cadence, NVIDIA Expand AI & Accelerated Computing Partnership

Estimated reading time: 5 minutes

At CadenceLIVE Silicon Valley 2026, Cadence announced an expanded partnership with NVIDIA to deliver accelerated solutions across agentic AI, physics-based simulation and digital twins to unlock new levels of productivity and accelerate next‑generation engineering design flows across semiconductor design, physical AI systems and hyperscale AI factories.

By combining Cadence’s leadership in agentic AI-driven design, electronic design automation (EDA) and system design and analysis (SDA) with NVIDIA CUDA-X, AI physics and Omniverse libraries for industrial digital twin solutions, the two companies are redefining engineering productivity across three critical design domains—accelerating innovation at true agent speed.

“Agentic AI and digital twins are reshaping the entire engineering landscape—from semiconductor design to planetary‑scale AI systems,” said Anirudh Devgan, president and chief executive officer, Cadence. “Our expanded collaboration with NVIDIA accelerates the convergence of design and physical realization, connecting the Cadence AgentStack, Physical AI Stack, and AI factory digital twins with NVIDIA’s breakthroughs in accelerated computing to deliver unprecedented speed, accuracy and trust in simulation and system development.

“We are at an inflection point in computing—CUDA-accelerated computing and AI are reinventing the engineering process,” said Jensen Huang, founder and CEO of NVIDIA. “For the first time, we can innovate in the digital world—exploring, testing, and optimizing ideas at unprecedented speed and scale—by building everything as full-fidelity digital twins first. Together, NVIDIA and Cadence are bringing this vision to life—transforming how engineers design, build and operate the world.”

Accelerating Cadence Tools for EDA and SDA

Cadence and NVIDIA are accelerating Cadence EDA and SDA solutions with NVIDIA CUDA-X, AI-physics, Omniverse libraries and the Cadence® Millennium™ M2000 Supercomputer, powered by NVIDIA AI infrastructure. As part of this expanded collaboration, Cadence will accelerate its wide range of principled solvers and leverage AI physics models to deliver engineering workflows up to 100X speedup.

Cadence EDA and SDA customers and partners, including Ascendence, Argonne National Laboratory, Honda R&D, Samsung and SK Hynix are already leveraging Cadence solutions accelerated by NVIDIA to bring accelerated products to market faster.

AgentStack: Agentic AI for Next-Generation Chip Design

Cadence recently introduced its ChipStack™ AI Super Agent, which applies agentic AI combined with principled EDA tools to transform semiconductor RTL design and verification. Early deployments at more than 10 leading customers have already demonstrated up to a 10X productivity boost in their design and verification tasks.

Building on this foundation, Cadence today unveiled AgentStack™, a head agent designed to orchestrate all aspects of semiconductor and system design. AgentStack extends the ChipStack AI Super Agent’s Mental Model and super-agent architecture beyond RTL and verification into physical design, custom/analog design and migration, to system-level design workflows. AgentStack connects Cadence agents with Cadence EDA platforms that leverage NVIDIA Nemotron and run on NVIDIA accelerated computing for orchestrating long‑running, multi‑agent workflows.

As an early partner, NVIDIA is adopting the AgentStack flow in its semiconductor and system design flows and providing real-world feedback that will help Cadence harden and scale AgentStack for broader industry deployment. This evolution will mark a significant shift from traditional script‑ and GUI‑driven flows to agent‑driven flows that are capable of reasoning over design hierarchies, relationships and protocols, dramatically compressing iteration cycles from days to hours.

Embedded Agentic AI for Physical AI

Beyond semiconductor design, Cadence and NVIDIA are extending their collaboration to embedded agentic AI for physical AI, combining the Cadence Physical AI Stack with NVIDIA robotics simulation libraries and accelerated computing to help close the critical “sim‑to‑real” gap for robots and autonomous machines. By integrating and accelerating Cadence’s high‑fidelity multiphysics simulation and AI workflows with NVIDIA Isaac open-source simulation libraries and Cosmos open-world models, customers gain an end‑to‑end, agent‑orchestrated workflow that links world‑model training, accurate physics, large‑scale scenario testing and continuous real‑world feedback.

At a high level, the joint stack coordinates AI agents across the full lifecycle—from training orchestration, physics surrogate training and policy optimization, to validation and deployment feedback. This workflow spans virtual training in NVIDIA Isaac Sim and Isaac Lab, evaluation through detailed Cadence physics models and mission‑scale scenario simulation in VTD (Virtual Test Drive) and VTDx—its extended high‑fidelity simulation environment for complex, real‑world scenarios.

The results are then deployed on NVIDIA Jetson robotics and edge AI systems, where a live virtual twin enables continuous monitoring and refinement. By embedding accurate physics throughout training, validation and inference, the Cadence–NVIDIA flow is designed to greatly accelerate experimentation while improving safety and confidence when physical AI systems are deployed in the real world.

AI Factory Digital Twins to Achieve Lowest Cost per Token

The collaboration also extends to AI factories, where Cadence integrates the NVIDIA Omniverse DSX Blueprint to enable next-generation AI factory digital twins that will help customers design, simulate and optimize large‑scale Vera Rubin and Grace Blackwell AI factories for training and inference. These AI factory digital twins focus on a critical new metric for hyperscale AI: tokens per watt, or the number of model tokens processed per each unit of power consumed.

Using Cadence system analysis and data center simulation tools in combination with NVIDIA DSX libraries and the Omniverse DSX blueprint, customers can explore tradeoffs in GPU power settings, system configurations and cooling architectures before deploying physical systems. In a joint 10-megawatt (MW) AI factory use case, modeling GPU operation at a reduced power (MaxQ) demonstrated up to 17% more tokens per watt and billions of dollars of incremental annual revenue per gigawatt for large‑scale deployments, increasing net annual revenue and underscoring the value of simulation‑driven design for AI factories.

Digital twins of NVIDIA DSX-based AI factories have also demonstrated that combining MaxQ operation with warmer coolant could yield roughly 32% more tokens per watt. By capturing the interactions between IT load, cooling systems, airflow and control logic in a high‑fidelity digital twin, operators can safely push their AI factories toward maximum tokens per watt while respecting power and thermal constraints.