this The interview analysis is sponsored by Microsoft and Nvidia. It was written, edited and published in an alignment with us Content guidelines sponsored by Emerj. Find out more about thought leadership and content creation services on the Emerj Media Services page.
Manufacturers around the world are increasingly pressured to meet evolving market demands and improve operational efficiency and agility.
According to the National Institute of Standards and Technology (NIST), manufacturers are increasingly relying on operational dashboards to monitor key performance indicators (KPIs) in real time, enabling proactive maintenance, throughput optimization, and quality control. Dashboards are at the heart of smart manufacturing initiatives, providing visibility across different systems and improving responsiveness at the field level.
McKinsey estimates that digital conversion in manufacturing can improve labor productivity by 15% to 30%. We estimate that this increased productivity can be affected through manual task automation, increased operational transparency, and decisions to make AI support.
However, the transition from manuals to digitized systems presents new challenges. Migrating to a simulation dashboard requires standardized data entry, scalable infrastructure, and interoperable systems that can support real-time edge computing.
What exacerbates these challenges is the fact that many manufacturers face sustainable data barriers. Research from Data Integrity Leaders accurately found only 12% of organizations reporting to have sufficient quality and accessibility data to support effective AI implementation, in collaboration with the Applied AI and Business Analytics Center at Drexel University's Lebow Business College (Drexel Lebow) .
Meanwhile, 64% cited data quality as the greatest data integrity challenge. This is a significant increase from 50% in 2023. These data requirements and broad gaps in integrity and accessibility highlight the current difficulty driving AI integration.
Matthew Demello, editor-in-chief of Emerj, recently held a conversation with Rad Desiraju, corporate vice president of manufacturing at Microsoft, and Mike Geyer, director of industrial AI at Nvidia.
Their discussions addressed topics such as data interoperability, infrastructure scaling, and the practical benefits of simulation dashboards. Both Rad and Mike highlighted the importance of GPU-accelerated edge computing in promoting integrated platforms and operational efficiency, safety and predictive capabilities in manufacturing.
This article explains two important insights from manufacturer conversations.
- Promote manufacturing intelligence through 3D digital twins: The transition from traditional monitoring dashboards to a generic, AI-driven 3D digital twin allows real-time simulation analysis to improve decision-making for improved throughput, safety and operational agility.
- Building a scalable, interoperable infrastructure: Successfully deploy AI by standardizing a variety of data sources, adopting containerized edge computing, leveraging GPU-accelerated platforms to reduce latency, improve data interoperability and drive value.
Listen to the complete episode below:
guest: Rad Desiraju, Director of Microsoft's Worldwide Industry Advisory
Expertise: Manufacturing, Digital Transformation, Industry Advisory
Simple recognition: RAD is a global industry advisor to Microsoft's manufacturing industry, helping to adopt next-generation solutions such as AI-powered Digital Twins and Edge AI. He is a presenter focusing on interoperability and standards at events such as Semicon West and Semi's “Digital Twin” workshop. RAD is a recognized thought leader who actively advises manufacturers on operational modernization through collaboration with industry partners.
guest: Mike Gayer, head of Nvidia's digital twins
Expertise: Digital Twin, Industrial AI, Robotics, Platform Strategy
Simple recognition: Mike leads Nvidia's Industrial AI Initiative, promoting the development and adoption of digital twins, employing simulation technology and libraries for manufacturing. His previous roles in Caterpillar and Autodesk bring him decades of domain experience. His recent LinkedIn posts highlight NVIDIA's collaboration with major manufacturers, transforming industrial and physical AI.
Promote manufacturing intelligence through 3D digital twins
Rad Desiraju outlines three distinct stages to set the stage for the evolution of manufacturing dashboards over the last few decades. The way he explains this history serves as a spectrum of where many manufacturers stand across the global economy, in terms of the refinement of dashboard development.
- Diagnostic dashboard showing what happened
- Operational dashboard showing what's going on
- A simulation dashboard that allows manufacturers to investigate potential outcomes under a variety of conditions.
Rad emphasizes that while many manufacturers monitor dashboards to visualize factory data, most remain stuck in the first two stages, unable to simulate real-time “What-if” scenarios.
This is because meaningful simulations (the kind that help leaders make positive decisions) require access to reliable, standardized data and infrastructure that can interpret them with high fidelity.
“One example of a development platform is something we are working on at Omniverse. This allows us to use OpenUSD to gather data from various sources and actually combine this collaborative environment in 3D.
We have dreamed of what we have been talking about for decades at a time when technology, acceleration, data platforms and analytics are coming together. It feels like the pace of change is really accelerating quickly. ”
– Rad Desiraju, Director of Microsoft's WW Industry Advisory
Mike Geyer has expanded this concept, noting that most product manufacturers have already used 3D product designs since the 1990s. The next step is to apply the same level of dimensions and analysis to the factory floor.
According to Mike, the simulation dashboard allows you to dynamically model the entire production environment. Adjusting product mix, rerouting supply lines, optimizing material staging, etc. – no physical trial and error:
“The power of the backend is something that goes very quickly as computing becomes scalable in the cloud. Manufacturing facilities are not 2D. When you walk around, you may hit your head on something that is not visible on a 2D dashboard.
These are tall, 3, 4, or 5 stories with complex dependencies, where things move horizontally around the floor, as well as up and down. When training physical AI, there is a physical world, and you need to be able to simulate it. So, GPU acceleration and open development platforms, and the intelligence that Microsoft can bring – sewing all of this together helps change that situation very quickly. ”
– Mike Geyer, head of digital twins at Nvidia
As Mike explains, the ability to effectively test and optimize factory performance is changing manufacturing decisions. Instead of producing based on forecasted demand, companies can now pursue just-in-time production strategies driven by real-time simulations.
Microphone further highlights how the rise of AI and automation, particularly the rise of humanoid robots and AMR (automatic mobile robots), can enable these changes by helping manufacturers model not only product flow but human workflows. Safety is a critical outcome, especially in high-risk environments where robots can assist with dangerous or repetitive tasks.
Building a scalable, interoperable infrastructure
A central insight from both speakers is that implementing AI-powered simulation and digital twin systems is not just a software upgrade, but an infrastructure challenge that spans the cloud, edge and physical factory environments.
RAD provides an overview of the three different computing environments needed to ensure a successful deployment. One trains an AI model, one simulates digital twins, and the other performs inference at the edge. Each environment has its own computational and latency requirements to generate real-time insights, and needs to be seamlessly adjusted.
“The way to address this issue is to try and look into three vertical archetypes. The first archetype is to build a modern data architecture, or something called a “data lake.”
All structured data is put together and sets the foundation for talking to it and asking questions about natural language data. That was the first pillar we created. It is called structured data extraction. The second is called document intelligence. The third is making them interoperable. ”
– Rad Desiraju, Director of Microsoft's Worldwide Industry Advisory
Edge computing in particular plays an important role. Both speakers emphasize that for digital twins to be effective, manufacturers need to reduce latency and keep their data as close to the source as possible. This means adopting containerized infrastructure and GPU acceleration in both the cloud and on-premises to manage computationally intensive workloads such as 3D simulation and sensor data fusion.
Mike points out that a key differentiator for the Nvidia-Microsoft partnership is its ability to perform GPU performance right-sized performance for each specific workload. This flexibility helps manufacturers avoid excessive aid, reduce costs and reduce value times, especially in industries where minutes of downtime is millions.
“One of the things I've actually worked on with partners like Microsoft is how to make computing scalable through these open development platforms. This allows the development ecosystem to build digital twins with their own tools, augmented by some of the parallel computation and accelerated GPU capabilities.”
– Mike Geyer, head of digital twins at Nvidia
Mike explains that current practices of overall manufacturing are possible only with the right infrastructure, including updating warehouses every few years.
Next, RAD reinforces that conversations about these levels of digital conversion and the infrastructure needed to achieve them must be launched with clear business outcomes in mind. Having a clear and systematically important objective as a focus for AI adoption is important for a wide range of manufacturing use cases, including improving safety, increasing throughput, and reducing operational costs.
From there, teams can adjust data standards, computing needs and platform architecture accordingly. Rudd said, “The first rule is: Yes, the Digital Twins are beautiful, but the important thing is to ask questions. What business value are you trying to solve?”
