Whether in electric arc furnaces, heat treatment plants, or site-wide energy management — applications are emerging everywhere that analyze process data in real time and automate controls that used to rely on experience and fixed setpoints. At the same time, it is becoming increasingly clear that AI itself consumes energy, sometimes considerable amounts.

For the thermal processing and metals industry, one of the most energy-intensive sectors of all, it is worth examining both sides of this development.

Process level: Where AI engages directly with the furnace

The steel industry offers a vivid example, as it is responsible for a significant share of global CO₂ emissions — most of which originates in the classic blast furnace process through the use of coke as a reducing agent. The electric arc furnace (EAF) is considered a markedly lower-emission alternative, since it runs predominantly on electricity rather than coal and can flexibly process scrap, direct reduced iron, or liquid pig iron.

A decisive factor for the energy efficiency of an arc furnace is temperature control: the more precisely the melt temperature is known, the less energy the process requires. Traditionally, a so-called sacrificial sensor is inserted into the melt for this purpose — but it is destroyed in use and its deployment requires shutting down the electrodes, causing an efficiency loss with every measurement. ABB has developed an AI model that predicts the melt temperature non-invasively: based on the heating of the cooling water, the trained model continuously calculates the temperature inside the furnace without interrupting the melting process. The approach not only saves on sensor consumption but also enables tighter process control, which directly affects energy use and emissions. According to ABB, the company is working to integrate this method as an option into its own control technology platform.

The Fraunhofer Center for High-Temperature Lightweight Construction (HTL) is pursuing a similar path: there, digital furnace twins are being developed that model the interaction between the material being heated and the furnace itself. This makes it possible to mutually optimize process parameters and furnace design — with the dual effect of increasing energy efficiency while reducing scrap rates. Control concepts that also incorporate AI algorithms are first developed and tested virtually on the digital twin before being deployed in the real plant. From accompanying energy flow analyses, HTL also derives concepts for heat recovery.

Plant level: Digital twins and predictive maintenance

For large-scale new installations, the digital twin has become practically standard. SMS group has provided a digital twin, based on its own Genius CM and DataXpert systems, for SSAB’s new 190-ton electric arc furnace at the Oxelösund site, serving as a central interface for predictive maintenance. The furnace itself processes fossil-free direct reduced iron (DRI) or hot briquetted iron (HBI) alongside scrap and is part of SSAB’s strategy to achieve near fossil-free production by around 2030.

Process monitoring is also moving further into focus in modernization projects for existing plants. In June 2026, Tenova received the order for the revamp of an electric arc furnace at Tenaris’s plant in Koppel, Pennsylvania. At its core is a new system for precise monitoring of cooling water flow at the upper furnace shell, intended to provide greater stability and optimized process conditions — an example of how advanced sensor technology and data analysis are now being incorporated even into projects focused purely on efficiency and reliability, without every initiative being explicitly marketed as an “AI project.”

System level: From furnace to site

The third level at which AI comes into play is not the individual process, but the interplay of entire sites. Research company etalytics, for instance, links utility technology, production planning, and energy forecasting in its ENIPRO project, tested at TU Darmstadt’s ETA Factory, to reduce costs and emissions across sites. Another of the company’s projects aims to minimize power usage effectiveness in cooling systems using real-time data and digital twins, promising energy savings of up to 50 percent.

A recent survey of industrial companies on their 2026 efficiency priorities shows that this systemic perspective is increasingly becoming the norm: providers such as Copa-Data are focusing on the end-to-end capture and integration of production, energy, and building data — from the field level to the management level — to make consumption visible, control loads intelligently, and systematically identify waste heat recovery potential. Emerson, too, is expanding its automation platform with a GenAI-capable component. The common thread: energy efficiency is increasingly understood not as the sum of individual measures, but as the result of comprehensive transparency across all energy and process data.

The counterpoint: AI itself consumes energy

What is often underemphasized in many efficiency reports is that AI applications themselves are energy-hungry. According to figures from the Borderstep Institute, energy consumption by data centers in Germany could more than double by 2030, with AI workloads potentially requiring up to ten times more energy than conventional IT applications. For manufacturing companies looking to introduce AI-supported automation, IT infrastructure itself thus becomes a challenge — additional cooling capacity and data center space must be planned for, which can delay implementation if existing infrastructure is not adequately sized.

A recent PwC market study on AI in the energy sector puts this development into context: in 2026, AI in this sector crossed the threshold from experiment to productive value driver, with roughly a third of surveyed companies citing efficiency gains from automated processes and lower operating costs as the greatest benefit. Nearly 60 percent of companies expect AI to become a strategic or even transformative element within the next five to ten years — with the trend moving away from isolated automation toward intelligent, cross-process orchestration.

Conclusion

A nuanced picture is thus emerging for the thermal processing and metals industry: at the process level, AI delivers measurable efficiency gains, for instance through non-invasive temperature measurement in the electric arc furnace or digital furnace twins that jointly optimize furnace design and energy efficiency. At the plant level, digital twins for predictive maintenance have become standard in new builds, while modernization projects are increasingly relying on more precise sensor technology and data analysis. And at the system level, linking production, energy, and building data promises savings potential that individual measures alone cannot achieve. The AI infrastructure’s own energy appetite, however, tempers this optimism.

Sources

• ABB / Destination Zukunft: “Steel melting in the electric arc furnace: even more efficient thanks to ABB AI!”

• PwC: “AI in the Energy Industry 2026”

• Deutsche Rechenzentren GmbH: “Energy Challenges in the Age of AI”

• industr.com: “Where are the greatest efficiency potentials in industry in 2026?”

• etalytics: “AI Research and Innovation”

• Oekologisch Erfolgreich: “Thermal Process Optimization 2026”

• Fraunhofer-HTL: “Thermal Processing Plants”

• SMS group: “SMS group builds a new electric arc furnace for SSAB”

• Tenova: “Tenova to Supply Electric Arc Furnace Revamp for Tenaris’ Koppel Steel Mill”