Summary
The episode digs into Siemens’s evolution from a collection of hardware-focused divisions into a platform-oriented ‘‘one tech company’’ that blends industrial hardware, software, and services across many verticals. Central themes are digital twins, AI (including LLMs as components), and closed-loop ‘‘industrial agents’’ that ingest real-time machine and product data to monitor, simulate, and act on manufacturing lines. The conversation covers organizational change (horizontal ‘‘fabrics’’ for data, technology, and sales) to scale AI advantages, plus the geopolitical pressures (localization, tariffs, semiconductor dual-sourcing) reshaping manufacturing footprints. Practical engineering details — photorealistic sim-to-real training, the limits of generic LLMs, and the need for high-quality, domain-specific datasets — illustrate how Siemens aims to reach industrial-grade reliability while balancing customer data ownership and trust.
Key Takeaways
- 1Siemens is transforming into a platform-style tech company that pairs industrial hardware with shared software and data fabrics.
- 2Digital twins plus orchestration agents create closed-loop automation that augments human operators on the shop floor.
- 3Generic LLMs are insufficient for industrial automation; domain-specific, high-quality data and augmentation are required.
- 4Photorealistic simulation materially improves sim-to-real transfer for robotic tasks.
- 5Geopolitical fragmentation is forcing localization and platform forking decisions that trade scale for resilience.
- 6Customer data ownership and trust are central to industrial AI scale-up; data alliances are a governance path forward.
Notable Quotes
""Every third manufacturing line in that world is run by Siemens controls.""
""I think we are controlling, I mean, something like a little bit less than 50% of electrons are touched by Siemens technologies...""
""We still invest out of our 8% in terms of revenue, or $6.5 billion.""
"Then we used NVIDIA technology with a photorealistic ray tracing of these pieces."
"The hit rate was still not satisfying, 70 odd level... Photorealistic ray tracing... hit rate was jumping up substantially."
"All that data has to come from lots of different customers... but that data actually belongs to your customers."
"We are training our industrial eye applications for China on Chinese LLMs, whereas for the United States obviously we train them on American hyperscalers."