Physical AI enables systems to perceive changing conditions, reason over context, and act in the physical world within defined boundaries. For automated-driving teams, developing those capabilities starts with a practical engineering challenge aimed at turning driving data into evidence they can use to evaluate software. The work with Porsche brings that challenge into focus, with Microsoft, NVIDIA, and Voxel51 supporting a customer-led approach to AV development.
Microsoft’s vision for Physical AI connects the full lifecycle, from data and simulation through training, evaluation, deployment, and operational feedback. Our partnership with NVIDIA brings complementary capabilities into that workflow, as Microsoft Azure provides compute, data services, orchestration, and governance, while NVIDIA contributes physical AI technologies for data curation, retrieval, synthetic data generation, and scene reconstruction. Together, we aim to help engineers build a repeatable development process, with customers retaining control of their data, engineering decisions, and validation.
For me, the strategic opportunity is how each development cycle improves the next. A driving scenario captured today can become reusable evidence across vehicle programs and software generations. That is where this work can create lasting value for automakers.
The future of automotive software
The next competitive advantage in automotive software will come from how effectively an organization turns each mile into reusable validation evidence. That evidence should serve the vehicle being developed today and the generations that follow. Collecting road data is a starting point. The strategic question is whether teams can use what they learn to make better engineering decisions, repeatedly and across programs.
This is the promise of autonomous vehicle operations, or AVOps: an engineer-directed development loop that connects continuous road data, curated scenarios, simulation, model improvement, validated software, and real-world feedback. It supports the development of autonomous driving and advanced driver assistance systems, or ADAS. Engineers determine what needs to be learned, choose the evidence that matters, and retain authority over validation and release decisions.
At Microsoft, we are working with NVIDIA and automotive partners to make that loop more connected and practical. Three priorities shape my view of what it takes: purposeful data, repeatable development, and customer control. Together, they create a foundation for learning that can compound across software releases, vehicle programs, and brands.
Turn driving data into engineering evidence
A development team needs to find the scenarios that expose a gap in model behavior, understand their quality, and turn them into useful tests. Search, scenario mining, annotation, and curation give road data a specific engineering purpose. Synthetic generation can then help expand the scenarios available for development, with engineers setting the questions and assessing whether the resulting data is suitable.
The ecosystem brings complementary capabilities to this work. NVIDIA is bringing its technologies to Azure, including NVIDIA Cosmos Data Factory for search and synthetic generation and NVIDIA Omniverse NuRec for scene reconstruction. Voxel51 integrates with Azure and uses these NVIDIA technologies alongside its capabilities for data quality validation, search, scenario mining, annotation, and curation.
Azure provides compute, storage, orchestration, integration, and governance to connect these capabilities into workflows that automotive teams can operate.
For me, the value of this foundation is the connection between a development question and the evidence used to answer it. A scenario becomes more valuable when another team can understand why it was selected, reuse it, and compare results. That is how learning from one program can become an asset for the next.
Connect development around repeatable evidence
Connected workflows allow engineers to carry a scenario from road data into reconstruction, simulation, and model improvement, then bring the results back into validation. Shared workflows also make scenarios reusable across programs, releases, and brands. The opportunity is to build a growing body of evidence so teams can revisit a known condition when software changes and see whether behavior has improved or regressed.
Virtual validation compares reconstructed camera inputs and simulated outputs with real measurements. The aim is to enable more repeatable testing, faster feedback, and fewer repeated drives for tests that can be meaningfully reproduced. Simulation has limits: it does not establish road safety or replace all real-world testing. Engineers must assess how closely the virtual conditions represent the physical ones and decide what further evidence is required.
I would measure speed through the validated release. Generated data volume alone tells us little about engineering progress. The useful questions are whether a workflow helps resolve a development issue, whether its results can be reproduced, and whether the evidence supports the next decision. Azure’s role is to support the compute and operational connections that help teams move through that process.
Maintain OEM control across AV development
Porsche, working with Volkswagen Group brands and CARIAD, is building an OEM-controlled, modular approach to a shared autonomous vehicle data flywheel. Porsche is playing a key role in the development of a shared AV data flywheel across Volkswagen Group brands, using Cosmos Data Factory through Voxel51 on Microsoft Azure. This collaboration is expected to reduce repeated physical test drives by up to 90% and shorten AI regression test cycles by up to 85%. The direction is significant: shared capabilities and reusable scenarios can support collaboration while keeping the automotive manufacturer in control of its development process.
That control is central to trust. Customers need to govern their data, choose how capabilities fit together, and maintain responsibility for their engineering decisions. Microsoft’s intelligent, trusted platform with a broad partner ecosystem gives customers the flexibility to bring specialized tools together around their own business priorities securely and at scale. A modular ecosystem gives customers a way to connect specialized tools around their own priorities.
My view is that the lasting opportunity extends beyond any single tool: each development cycle should leave behind evidence that the next one can use. When engineers can turn road experience into repeatable validation and carry that learning across vehicle generations, the value of every mile can continue to grow.
Check out these resources
- Explore the NVIDIA AVOps on Azure Reference Implementation.
- Learn more about Microsoft mobility solutions.
