In a new Siemens factory that makes large switchboards for data centres in Fort Worth, Texas, artificial intelligence is doing a lot of work.
The demand for switchboards is surging as companies build out AI foundation models that power this new technology, but Siemens wants to do more than supply widgets for AI; the German manufacturer also wants to provide AI infrastructure.
The Munich-based company announced earlier this year that it would build its own foundation model that’s focused on industry. This pits Siemens against companies spending billions on AI, such as Google and Nvidia, which are betting their massive models can be fine-tuned to specialise in automation and industrial applications. Although the capabilities are underpinned by the big model, the fine-tuning cuts out the clutter, bad information or hallucinations to make the models safe and efficient for industrial use.
Siemens’ selling point is that the German company knows industry because it makes things, including robots and gas turbines. It also knows software. More than two decades ago, Siemens and other industrial automation companies such as Honeywell International and Emerson Electric pivoted toward adding more software and data to their products and services.The key will be persuading customers — fellow industrial companies — to contribute their data to the Siemens pot that trains the foundation model.
Siemens isn’t saying much about the creation of this model, but the pitch to customers is that the industrial focus will provide the engineering know-how, operations safety and data security that manufacturers require.
On its webpage that invites industrial customers “to join our journey”, Siemens says: “To meet the rigorous demands of industry, it must be able to understand the language of engineering. That’s why we’re developing an Industrial Foundation Model that takes AI in industry to the next level.”
Tolerance for errors
The established AI companies and start-ups, such as OpenAI and Anthropic, are already revolutionising office tasks — doing research, taking notes, coding, accepting customer orders or making memes. The tolerance for errors, for now, is much higher in the office setting.
Will the tech companies eat up the industrial AI market or will industrial companies with software experience triumph with deeper knowledge of manufacturing? Could this same dynamic play out in other sectors, such as legal or transportation?
A similar showdown of big tech versus established players occurred in trucking when digital freight brokers arose out of Silicon Valley to revolutionize the industry. In the end, the established brokers adopted the tech and held onto their customers while most of the digital-only brokers went bust.
Read: AI to replace line judges at Wimbledon
The AI competition isn’t as black-and-white because of all the partnerships being formed. Microsoft has partnered with Siemens and other industrial companies including Honeywell. Siemens has a partnership with Nvidia and is tapping Microsoft’s Azure platform to build its foundation model.
With manufacturing, safety comes first and it’s not good for business to interrupt production. That’s why Siemens is being careful about how it introduces this technology at its Fort Worth plant.
Even before Siemens installed the equipment at the factory, a digital twin of the factory operations helped engineers try out different configurations of the machines and supplies to optimise the flow of sheet metal, copper, wires and other materials to make large switchboards. The digital twin also monitors operations in real time, providing data that can uncover bottlenecks and can be monitored remotely.
QR codes match the materials that are pulled from storage shelves and placed on carts with one of the 35 different designs that are made at the plant. Siemens is working on making the carts move autonomously. The assembly procedure is mapped out on a computer screen, so workers don’t have to memorise all the designs or fumble though paper blueprints.
In a laboratory within the same cavernous 80 000sq m facility, engineers are testing a robotic arm with an attached camera that hovers over subassemblies to inspect for quality.
Siemens also plans to use automated impact drivers that tighten bolts to exact torque specifications. Those are now done by hand and then measured for the proper tightness. Siemens’ AI system also captures data from the machines that cut, bend and paint sheets of metal to form the cabinets. If a machine isn’t performing as it should, the AI will flag the problem. Modern factories shed a lot of data. With its foundation model, Siemens is throwing its hat in the ring of tech competitors seeking to collect it and to provide the AI services to industry.
In the end, AI is a tool that should be designed to help workers be more efficient. Siemens has simplified assembly jobs, enabling workers to hit the factory floor after about three weeks of training. More than a quarter of the more than 500 workers hired so far at the Fort Worth factory had no previous manufacturing experience.
Boundaries
As robots, especially mobile ones, become more prevalent in the workplace and society, it will become even more important to limit the AI capabilities to the tasks at hand. Putting boundaries and restrictions on the AI will keep these machines operating safely around humans.
A know-it-all AI research assistant that sometimes goes off the rails by including something goofy in a slide presentation may be fine for office work. On the factory floor, those mistakes are unacceptable. That could give Siemens the upper hand in AI. — Thomas Black, (c) 2025 Bloomberg LP
Get breaking news from TechCentral on WhatsApp. Sign up here.
Don’t miss:
Naspers shifts to an AI-first strategy – and it’s paying off