ASOTECH BLOG
NVIDIA GTC 2026 and physical AI: the CAD-simulation-factory bridge is real. What’s changing for the engineering department
At GTC 2026 NVIDIA transformed physical AI from a concept to an industrial infrastructure. FANUC, ABB, KUKA and Yaskawa integrate Omniverse into their systems. PTC launched a direct link from cloud CAD to physical simulation via OpenUSD. For those who design and manufacture, the design-simulation-deploy flow becomes seamless. Those who do not adopt it risk designing in the past.
Two million industrial robots installed worldwide. Four manufacturers (FANUC, ABB, KUKA, Yaskawa) that together cover the majority of the global base. At GTC 2026 in San Jose, all four announced the integration of NVIDIA Omniverse and Isaac Sim into their development and validation systems. This is not a generic partnership: it is an infrastructure change.
The most quoted sentence of Jensen Huang’s keynote was straightforward: every industrial company will become a robotics company. Provocation? Perhaps. But when PTC announces a native bridge between Onshape (cloud CAD) and Isaac Sim (physical simulation) through OpenUSD, the provocation becomes workflows. And workflows change engineering departments.
This article looks at what happened at GTC 2026, why physical AI is no longer a conference slide, and what it means concretely for machine, line, and plant designers.
What is physical AI and why has it become an industry fact at GTC 2026?
Physical AI is the set of models, simulations and frameworks that enable physical machines (robots, vehicles, plants) to perceive the environment, make decisions and act in the real world. Unlike generative AI, which produces text or images, physical AI moves objects, assembles components, and navigates spaces.
Until 2025 it was territory of spectacular demos and limited pilots. At GTC 2026, the roster changed. NVIDIA presented GR00T N1.7, a foundational model for robots now available under commercial license. Not a prototype: a product with pricing and support. In parallel, the Physical AI Data Factory Blueprint offers an open architecture for generating synthetic training data at industrial scale, solving one of the most stubborn bottlenecks in robotic development.
However, the strongest signal comes not from NVIDIA itself, but from the ecosystem. When ABB, FANUC, KUKA and Yaskawa adopt the same simulation platform, the message is that physical AI has moved beyond the experimental stage. It has become infrastructure.
From Onshape to Isaac Sim: the CAD-simulation bridge that was missing
For those who work in an engineering office, the problem is familiar. You design in a CAD environment, export (with inevitable loss of data and formats), import into a simulation environment, validate, go back. Each step is an opportunity for errors, delays, and frustration.
PTC announced at GTC 2026 a direct workflow from Onshape to NVIDIA Isaac Sim through OpenUSD. Basically: the designer draws in cloud CAD, and the model arrives in the physical simulation without manual conversions. FANUC America and Fauna Robotics are among the first engineering teams to test this flow to design and validate robotic systems within physically accurate digital twins.
It is not a technical detail. It is a paradigm shift in the relationship between design and validation. The time between “I finished the model” and “I know if it works in production” compresses from weeks to hours. And it compresses without requiring the designer to become a simulation expert.
For those involved in industrial automation, the consequence is concrete: virtual validation of robotic cells and automated lines becomes part of the design flow, not a separate downstream activity.
FANUC, ABB, KUKA, Yaskawa: why did robotics bigwigs choose NVIDIA?
The four largest industrial robot manufacturers have a combined installed base exceeding 2 million units. At GTC 2026, all four announced the integration of NVIDIA Omniverse and Isaac frameworks into their virtual commissioning environments. The reason is practical: they need physically accurate digital twins to validate complex robotic applications before deployment.
ABB is integrating Omniverse into RobotStudio with a release HyperReality planned for 2026, designed to improve sim-to-real accuracy. FANUC is combining its systems with Isaac Sim, Omniverse and IGX Thor to accelerate intelligent automation and reduce setup times. On the AI side, Skild AI is collaborating with Foxconn for precision assembly on NVIDIA’s Blackwell chip production lines: double-armed robots that master high-complexity assembly tasks.
The point is not the individual partnership. It is the convergence. When the four global leaders in industrial robotics adopt the same simulation and AI layer, a de facto standard is created. For an integrator or engineering department designing automated lines, this means that virtual validation on Omniverse becomes the common language of the industry.
Digital Twin Composer and the PepsiCo case: +20% throughput before you hit a bolt
How much is a digital factory twin worth? Siemens and PepsiCo are answering with numbers. Siemens Digital Twin Composer, launched at CES 2026 and built on NVIDIA Omniverse libraries, enables the creation of a virtual 3D model of any product, process or plant, linked to real-time operational data.
PepsiCo is using it to transform some U.S. factories and warehouses into high-fidelity digital twins. Initial results speak for themselves: a 20% increase in throughput on early deployments, near 100 percent design validation, and a 10-15 percent reduction in capex by identifying hidden capacity in the virtual environment. All prior to any physical changes to the system.
For a manufacturing company, the message is that the digital twin is no longer a multimillion IT project reserved for multinationals. With platforms like Digital Twin Composer, it becomes an engineering tool: you simulate the changeover, test the new layout, validate the addition of a robot, then decide whether to proceed. Investment risk is reduced because decisions are made on virtual evidence, not estimates.
OpenUSD as an industry lingua franca: PLM needs to change
Behind all these announcements is a format: OpenUSD (Universal Scene Description). Born in the world of special effects (Pixar), OpenUSD has become the exchange standard for industrial digital twins. NVIDIA, Siemens, PTC, Dassault Systèmes and dozens of others support it.
Why is it relevant to an engineering department? Because OpenUSD is the first format that allows not only geometry, but also physics, materials, kinematic relationships, sensory data, and behaviors to be described in a single interoperable file. When PTC exports from Onshape into OpenUSD to Isaac Sim, it brings with it not only the part shape, but the simulation context.
The medium-term consequence is that PLM and PDM systems will have to manage native OpenUSD assets. Those who have built their product lifecycles on proprietary formats will have to integrate an interoperability layer that does not exist in standard workflows today. This is not tomorrow’s problem; PTC, Siemens and Dassault are already building bridges. The question for every engineering department is whether their PLM will be ready when those bridges become the norm.
Italy in the NVIDIA plan: AI factory in Europe and collaboration with Domyn
NVIDIA is not only looking to the United States. At GTC 2026, it announced the construction of the first industrial AI factory in Germany, designed specifically for European manufacturing, with 10,000 Blackwell GPUs in RTX PRO Server.
In Italy, NVIDIA is collaborating with Domyn to develop sovereign AI capabilities. Domyn is developing its reasoning model (Domyn Large Colosseum) on a supercomputer with NVIDIA Grace Blackwell chips, with the goal of supporting regulated industries in AI adoption. European telcos involved in the AI infrastructure include. Fastweb.
For Italy’s industrial fabric of highly specialized manufacturing SMEs, the availability of cloud AI infrastructure in Europe is a prerequisite. Until now, the bottleneck was not just software, but the computing power needed for physically accurate simulations. With European AI factories and local partners, that barrier is lowered. It remains to be seen how quickly Italian companies will be able to seize the opportunity.
What should an Italian technical office do tomorrow?
Physical AI does not require turning everything upside down overnight. It requires understanding where you are going and starting to move. Here are four concrete steps.
First, check if your CAD supports export to OpenUSD, or if it is on the vendor’s roadmap. This will become the main channel to physical simulation in 12-18 months.
Second, map processes that today are only validated with physical prototypes or spreadsheets. Robotic cells, changeovers, line layouts-these are ideal candidates for an initial digital twin pilot.
Third, start thinking about internal expertise. Physical AI does not eliminate the designer; it puts him or her at the center of a richer flow. But you need simulation foundations and the ability to interpret results.
Fourth, look for partners who know both the world of mechanical design and the world of simulation andautomation. The CAD-simulation-factory bridge only works if someone crosses it competently on both sides.
A change of scale, not direction
GTC 2026 did not invent anything radically new. Digital twins have existed for years. Robotic simulation also. OpenUSD was already circulating. What has changed is the scale: the four largest robot manufacturers in the world adopting the same platform, a CAD cloud natively connecting to physical simulation, a European AI factory dedicated to manufacturing.
For those who design and manufacture in Italy, the trajectory is set. The design-simulation-factory flow becomes continuous, and physical AI is the connective tissue. You don’t need to rush: you need to understand where you are today, where this direction leads, and what first step makes sense for your reality.
The Asotech team follows these developments closely. If you would like a discussion on how these trends may intersect your work, write to us at.
Frequently asked questions
What is the physical AI announced by NVIDIA at GTC 2026?
Physical AI is the set of models and frameworks that enable physical machines to perceive the environment, make decisions and act in the real world. At GTC 2026 NVIDIA released GR00T N1.7, a commercially licensed foundational model for robots, and the Physical AI Data Factory Blueprint to generate synthetic training data at scale.
What changes for those using CAD software such as Onshape, CATIA, or NX?
The main change is the direct integration between CAD and physical simulation. PTC has announced an Onshape → Isaac Sim bridge via OpenUSD.. Siemens and Dassault Systèmes are building similar bridges. The designer will be able to validate their model in a physically accurate environment without manual conversions.
What is OpenUSD and why is it important for the industry?
OpenUSD (Universal Scene Description) is an open format that describes geometry, physics, materials and behaviors in a single interoperable file. Originated by Pixar, it has been adopted as an exchange standard for industrial digital twins by NVIDIA, Siemens, PTC and Dassault Systèmes. For the engineering department, it means being able to pass data from CAD to simulation without loss.
Is the digital twin also affordable for manufacturing SMEs?
Yes, the threshold for access is lowering. Platforms such as. Siemens Digital Twin Composer on NVIDIA Omniverse and the arrival of European AI factories with dedicated GPUs for manufacturing reduce both infrastructure cost and setup complexity. The advice is to start with a circumscribed pilot: a robotic cell, a changeover, a line layout.
What skills are needed in the technical department to adopt physical AI?
You don't need to become a data scientist. You need to integrate simulation bases into the mechanical designer's background: understanding how to set up a digital twin, interpret results, and work with interoperable formats such as OpenUSD. Physical AI does not replace the designer; it puts the designer at the center of a larger flow that connects design, validation, and manufacturing.