Siemens

Siemens owns its factories. Most software vendors never get a shop floor for testing their own AI on their own numbers.

At the Erlangen electronics factory, productivity rose 69% and energy consumption fell 42% over four years, with AI deployed across more than 100 use cases. The World Economic Forum named it a Digital Lighthouse in October 2024. Siemens' own factory page puts a second number directly on AI. It says quality is up 60% from AI applications running on the shop floor.

Nanjing is the newer one. Siemens' motor control plant there runs more than 50 AI applications. Against a 2022 baseline, lead times dropped 78%, field failures 46% and time to market 33%. Productivity was up 14% by 2024. Commercial pressure explains the urgency behind that work. Siemens says the delivery window its customers would accept had shrunk from 45 days to 10, and the plant had to hit it. In January 2026 the WEF added Nanjing to its Global Lighthouse Network, Siemens' fifth site on that list.

Read those percentages carefully, because Siemens does not claim AI produced all of them. Both releases describe digital-native factories where AI is one contributor among many. What they do prove is that Siemens measures its plants closely enough to publish seven figures about them. Most of its customers could not do the same for their own plants.

Siemens' own tour of the Erlangen factory, hosted on its video platform.

CEO Roland Busch told investors in November that Siemens uses Google's AI to improve its code for software-defined automation, and Anthropic's Claude to refactor its own software. In his words, the AI cleans up and simplifies code.

The same speech gave the numbers underneath that work. Siemens has 30,000 software engineers running roughly 900 separate versions of development tools, and it is consolidating those down to a few dozen. Busch also said AI has cut tender applications for complex projects from weeks to hours.

The newest example sits outside engineering. On 15 September, Siemens said it has put AI agents on its own inbound sales desk. Siemens says it has gone from over 2,500 unqualified inbound leads a month to engaging 100% of them, across 132 countries, for 18,000 sellers. That is coverage rather than revenue, and Siemens is careful not to claim otherwise.

Those are company statements rather than audited metrics, and none carries a euro figure. They're still more specific than what industrial companies this size usually say about running AI on themselves.

Siemens' annual report for fiscal 2025 contains no quantified internal AI disclosure. Nothing in it attributes a productivity or adoption figure to AI inside Siemens. The sustainability statement has no AI-attributed energy saving either, despite Erlangen's 42%.

Busch gave the 175,000-user figure at the annual meeting in February, and the Google and Anthropic detail at a strategy event the previous November. The only euro figure Siemens attaches to AI is its planned spend of more than €1bn over three years. What that spending returns has never been published.

Every quantified AI claim Siemens has made sits outside the audited accounts. Two issues ago, Chevron put $92 million and 143,000 at-risk hours against its robotics work. Siemens gives no equivalent number for any part of its AI. Section 8 covers what a CXO should do with that gap.

Siemens reports digital business revenue of €9.4 billion for fiscal 2025 and says it will double that by 2030, around 15% a year. Software annual recurring revenue grew 11% organically in the quarter to reach €5.7 billion, and the Digital Industries segment posted an 18.7% profit margin in the same quarter.

Siemens labels none of it AI revenue, because it publishes no such line. Digital business is the closest proxy it discloses. That is the number to watch if you want to know whether the AI work is paying for itself.

The Industrial Copilot shipped inside TIA Portal, the engineering environment Siemens says 120,000 engineers can reach. Nobody had to adopt a new tool or learn a second interface. Siemens says it will deploy nine new AI copilots across its software, with Teamcenter, Polarion and Opcenter among the products named. In June it launched Intelligence Center X, one governed system connecting industrial data, workflows and AI agents.

That's the advantage a generic AI company can't copy. Siemens owns the software where the engineering happens, the installed base that throws off the data, the plants where it can test on itself, and more than 1,500 AI experts who understand what the data means.

Siemens' own motor plant at Bad Neustadt was still testing a production copilot in early access in February 2025. The published conclusion was that many use cases can be implemented. Nothing quantified has followed since.

A copilot alone never produces numbers like those, because they come from connecting systems rather than from one assistant. The buying motion is different, though. Siemens publishes no trial and no price here, so it starts as a conversation with your account team.

Siemens is serious about AI and still has no return figure to show for it. Four things follow.

Sort every AI number into one of four tiers before anyone presents it.

Anything that fits none of the four isn't ready for the room.

Paste this into whichever model your team already uses. Feed it one AI claim at a time and it returns the tier plus the question you should be asking.

Get more such prompts in the Prompting Playbook (free for you)