Joe Bohman on AI, Simulation and the Factory of the Future

Full announcement (unchanged)

At Realize LIVE Americas 2026, Joe Bohman, executive vice president of PLM Products, described how Siemens is bringing its software portfolio together around a comprehensive digital twin. His keynote explored five connected themes: software-defined products, the evolution of Simcenter, Intelligence Center X, the Factory of the Future and cloud-based software. This transcript follows Bohman’s presentation, also available below on YouTube, with minor edits for clarity. Building a comprehensive digital twin through connected software Good morning, everybody. I’m Joe Bohman , executive vice president of PLM Products. I want to talk about the product realization of the future. In 1992, the United States put together the Dream Team—the best basketball team ever. Michael Jordan, Larry Bird, Magic Johnson. And they won. But they didn’t win only because they had the best players. They won because they played together as a team. This morning, I want to show you what happens when our dream team plays together. For us, everything starts with the strategy Tony outlined yesterday: the comprehensive digital twin. “Comprehensive” is so important. Mechanical, electrical, electronics, software, engineering, manufacturing, operations and lifecycle intelligence. I’m going to talk about how we’re putting AI into everything we do, and how we’re making these capabilities available to customers. So many customers ask us, “How can we accelerate? If I’m an SMB or an enterprise, how can we have all these capabilities—and how can we have them quickly?” Siemens has made €25 billion in investments that support this strategy. But the point isn’t just the investment. It’s bringing all these pieces together. Last year, we talked a lot about Altair and Dotmatics. The key is that we’re playing together as a team. I want to show five examples of that: software-defined products, the new Simcenter, Intelligence Center X, the Factory of the Future and cloud. Software-defined products: solving the hardware-software integration challenge According to Design News, 52% of engineering managers say software-hardware integration is their biggest problem. Tony referred to it yesterday as the “integration nightmare.” Why is it so hard to integrate software and hardware? There are a number of challenges. There can be 30 different DevSecOps tools in the software toolchain. Customers tell us they have millions of tests to run, 30 functional-safety standards and more than 100 cybersecurity standards. It seems like there’s a new cybersecurity standard every day. Think about the software team’s world: DevOps tools such as DevStudio, Claude Code and Artifactory from JFrog, plus an application lifecycle management system to connect requirements planning to test management. At the other end, the PLM system handles system and configuration management. Between them are MiL, SiL and HiL: model-in-the-loop, software-in-the-loop and hardware-in-the-loop. They’re there because software has to be developed before hardware is ready. The industry created these approaches to address that timing gap, but they can be too slow and too late. Customers are asking for continuous verification. The software team needs to write software before the hardware is ready—before the silicon, boards and harnesses are available. The industry developed model-in-the-loop, software-in-the-loop and hardware-in-the-loop approaches to address that problem, but customers are asking us to move to continuous verification. They want software teams to get going before the hardware is available. Here’s how our products work together to help. Polarion orchestrates the software process. Capital and Xpedition simulate the boards. Simcenter simulates the physics of the model. PAVE360 lets customers simulate the silicon before it is available. Teamcenter manages compatibility between hardware and software. Let’s take an example. We’re adding an avoid-and-detect feature to a drone. Everything I’m going to show happens before the hardware, chips, boards and harnesses are available. The software team adds a new neural network in Polarion for avoid and detect and releases it. My team runs Polarion all day, every day. We start testing, with Jenkins orchestrating the tests. We’ve run 76,000 tests and found that 12 aren’t working. Often, the problem is timing. We need to be under 100 milliseconds on the network, but we’re at 500 milliseconds. We know that before the hardware is available. We can look at the system architecture in Capital Systems Architect : there’s a LiDAR, a GPU and the connections between them. Then we go into PAVE360. It can understand and run the software before the silicon is ready, and it shows us that when the silicon arrives, the system won’t be fast enough. We’re at 500 milliseconds instead of 100, and the GPU isn’t sized for this particular workload. So we modify the architecture in Capital Systems Architect and add a neural processing unit to help. Then we rerun the tests and see whether the software will work when the hardware arrives. The tests that were failing now pass. That’s an example of how we can work together as a team to solve the integration nightmare between hardware and software. Apptronik is a great example of a customer using this approach with its humanoid robot, Apollo. Apptronik has partnered with Google, and Google DeepMind is doing the software. It’s a success story of bringing software and hardware together. Simcenter: combining simulation, AI, multiphysics and HPC The next story is the new Simcenter. According to McKinsey, many manufacturing companies take 40 to 50 months to get to production, and 60% of that time is spent in physical prototyping. How can we help? The new Simcenter brings together what we had with Simcenter and Altair. Whether you’re working with fluids, structures, modeling and visualization, or 1D simulation, it’s there. We call it a market-defining convergence of simulation, high-performance computing (HPC) and AI. AI powers the new Simcenter in three ways: faster engines, smarter execution and trusted outcomes. We’re adding AI into our simulation engines, including geometric deep learning—training neural networks with the simulations we run. Copilots and agents can help simulation engineers work faster. And trusted outcomes include capabilities such as PhysicsAI Generate , which can generate a shape that meets your needs. Let’s look at a few examples. The first is PhysicsAI and an airbag from Continental. A nonlinear simulation that takes several hours using traditional methods can run in seconds with PhysicsAI—240 times faster. This is where simulation is headed: the ability to simulate at much higher performance. The second capability is multiphysics. Think about a car: wind moves over it and hits the mirror; the air wraps around and hits the glass; the glass vibrates, and you get acoustics. Those are different types of physics that come together for aeroacoustics. Or think about a washing machine: the fabric is a structure, and there’s heat and motion. That’s multiphysics. …we’re putting AI into everything we do… LG Vehicle Solution told us it needed four different solvers to model all the physics of a vehicle. Using OptiStruct , our multiphysics engine, we were able to bring those four solvers down to one. Some of those solvers were from competitors. With OptiStruct, LG Vehicle Solution was able to go 20% faster. Leonardo is another example. Helicopters have a lot of vibration. For a helicopter radar system, you need to bring vibration and electromagnetics together. OptiStruct and Feko can help solve that multiphysics problem. The third capability is GPU acceleration. Everything we’re doing in the new Simcenter is GPU-enabled. With BMW, our Star-CCM+ fluids and thermal solver helped deliver a simulation that runs three times faster and uses six times less energy with a GPU-based approach than with a CPU approach. Those numbers were already heavily optimized. And then there’s SimSolid : meshless simulation connected to design. An aircraft engine manufacturer was able to use it to go 180 times faster. The new Simcenter is a market-defining convergence of simulation, high-performance computing and AI. I didn’t even have enough time to get into all the HPC capabilities, which are outstanding. One more thing: for everyone who has Designcenter, SimSolid is now available through value-based licensing directly in Simcenter. We think Designcenter and Simcenter together are the future of engineering. Intelligence Center X: bringing industrial data together for AI The third story is Intelligence Center X . It’s part of our AI story—not all of it, but a big part. You’ll hear us talk about faster engines, smarter execution and trusted outcomes. We have tremendous physics, geometry and data engines, and we’re adding AI into those engines. We also have Siemens copilots and agents. When we say “copilots,” we mean Siemens copilots, not Microsoft Copilot. Intelligence Center X is especially important for trusted outcomes. I can’t tell you how many customers have told us that AI starts with data. If you have good data, you have good AI. If you have bad data, you have bad AI. That’s why Teamcenter and the data backbone are so important. We’ve added 150 capabilities—hundreds, actually—to the system, all based on that data backbone. So what is Intelligence Center X? It’s a one-stop shop for agentic enterprise development. It brings together four things: • Mendix for agentic development and orchestration. • Graph Studio and a graph database to put data in context. • AI Studio for machine learning, including statistical process control. • Industrial ontology to bring data together and run AI with it. Every CIO I talk to has a PLM system, an ERP system, a CRM system and an in-house system. Each contains different information: parts in PLM, vendors in ERP, customers in CRM and warranty information in an in-house system. The question is, how do we bring all that information together? Intelligence Center X uses an industrial ontology to connect it. Here’s an example. An airplane has a hydraulic problem with its landing gear. The engineering team has identified a change and is ready to release it. In Teamcenter Copilot, I can ask for help executing the landing-gear change. It finds impacted items—a task that could take hours or days—in seconds. Then I can ask it to find alternative parts with minimum supply-chain risk and minimum warranty issues. The system looks across the ontology and recommends alternatives using PLM, ERP and CRM data. It can also identify simulations I can run to check that a replacement part will work. Then we can use the new Simcenter and its AI capabilities to execute the change. GSK has a graph with 15 billion nodes in Intelligence Center X. That allows it to scale up and analyze supply-chain risk, reducing the process from hours to seconds. Gartner put it this way: “Siemens is the company to beat in manufacturing AI.” Factory of the Future: designing and operating AI factories The fourth topic is the Factory of the Future. We heard Steve from PepsiCo talk about some of these elements yesterday, and I’ll talk about them again. Nvidia has talked about an AI factory, with $50 billion to build a gigawatt AI factory. One of our customers involved in this is Foxconn. Foxconn builds AI factories and tests components before they go out. They have a test room and asked us to help design and operate it. …we’ve never had this much interest in an offering. Using Digital Twin Composer and Process Simulate , we can lay out the factory and understand how the room will work. We’re not going to look at just one layout, or 100, 200 or 300. We’re going to look at a thousand and find the best one. Foxconn found the best one, in a way similar to what PepsiCo showed yesterday. But we’re not done. Foxconn also asked us to help operate the test room. In Opcenter , we can see there’s a problem: it doesn’t have an input for power or heat. I can go into Copilot and ask to add those to Opcenter. We’re going to “vibe code” Opcenter. With a prompt, we can change it. This is AI-assisted coding. Mendix knows how to vibe code, and Opcenter knows Mendix. Put that together and it’s amazing. Based on the prompt, the system modifies Opcenter. We call it “vibe coding without the hangover.”