Deep.Meta’s physics-informed AI cuts steelmaking emissions and costs in pioneering trial

Deep.Meta, a spin‑out with roots at Imperial College London, has taken an early step toward decarbonising one of the most energy‑intensive links in modern supply chains by demonstrating that a physics‑informed digital twin can materially cut steelmaking emissions and costs.

According to the original report, the company’s Deep.Optimiser PhyX platform , which fuses real‑time sensor data, materials‑science models and machine learning , achieved close to a 10 percent reduction in emissions during trials at Spartan UK’s Newcastle plate mill and will move into a live pilot there. Spartan is Britain’s only steel plate producer; the UK steel sector delivered about £1.7bn in gross value in 2024 and remains strategically important to domestic manufacturing. Industry data shows global steelmaking accounts for roughly 9 percent of CO2 emissions, making efficiency gains in furnaces a high‑impact target for meeting climate commitments.

How the system works
Deep.Optimiser PhyX builds a physics‑based digital twin of reheating and finishing operations to predict slab temperature, optimise furnace scheduling and recommend operational adjustments that lower energy use. The platform can simulate hundreds of production cycles in hours rather than months, sharpening short‑term control and enabling operational decisions that both reduce fossil fuel consumption and improve yield.

According to the announcement from the company, the Manchester Prize , a UK government competition for AI tools for the public good , is supporting deeper physics integration in the system as Deep.Meta competes for a £1m final prize in March 2026. The company has raised £2.1m since 2020 to develop the platform.

Claims, corroboration and measured benefits
Deep.Meta’s founder and chief executive Dr Osas Omoigiade said: “Steel is one of the most important materials on which our society is built. Yet, its production generates 9 percent of all global CO2 emissions. We cannot reach net zero without solving steel’s climate impact. We are developing Deep.Optimiser PhyX to tackle inefficiencies that result in avoidable emissions, a crucial step in helping to decarbonise the industry. Through the Manchester Prize we have been able to integrate physics into our AI platform, which enhances its prediction capabilities further.”

Independent case material referenced by the company suggests more modest, process‑specific gains are also credible. A Made Smarter case study documents a CO2 saving of almost 5 percent in the reheating and finishing stages , based on a 200 kg CO2/tonne baseline for that part of the process , and an energy saving equivalent to 24 kWh per tonne of steel, with productivity uplift potential around 20 percent for an operation producing 2 million tonnes a year. Those figures align with broader sector experience showing that targeted digital optimisation frequently yields double‑digit percentage improvements in specific operations, while whole‑plant savings vary by process and feedstock.

Explainability and industrial adoption
Deep.Meta argues that coupling physical laws with machine learning eases the “black box” concerns that have slowed industrial AI adoption. As senior machine learning scientist Dr Kwangkyu Alex Yoo told reporters: “Today’s machine learning models often operate as black boxes, lacking fundamental principles that clearly link inputs to outputs. This creates significant resistance when industries attempt to deploy AI technologies in real production environments. Our physics based machine learning approach addresses these challenges by incorporating the underlying physical laws into both the training process and data generation. This leads to models that are more explainable and trustworthy, while enabling more reliable and robust decision making.”

Operational and policy context
Energy accounts for a substantial share of steelmaking cost; Spartan UK’s chief executive Michael Brierley noted that energy represents around 40 percent of production costs and much of it is fossil fuel based. “Deep.Meta is a trusted partner, and we are piloting the Deep.Optimiser solution, because of the rising costs of energy and carbon. Increasing the efficiency of production is of high importance as energy costs form a significant part of our cost structure. Around 40 percent of steel production costs are from energy and much of this is fossil fuel based, so driving a reduction in energy directly cuts CO2 emissions,” he said.

Trade‑exposed mills face competition from low‑price imports and tighter carbon policy; operators see digital optimisation as a way to preserve competitiveness by improving yield, product consistency and production stability while lowering emissions intensity. Industry bodies emphasise that technology and policy must advance together: Jon Bolton, co‑chair of the UK Steel Council, said collaboration between government and industry is vital, while Chris Oswin, chief executive of the Materials Processing Institute, framed AI as central to keeping UK steel competitive and resilient.

Implications for scale and investment
If physics‑informed optimisation can be replicated across furnaces and feedstocks, it could act as a pragmatic bridge to deeper decarbonisation routes , such as scrap‑rich melts, hydrogen reduction or novel low‑carbon steel grades under development with partners including academic institutions and major producers. For investors and plant operators, verified energy and CO2 reductions translate directly into lower operating costs and reduced exposure to carbon pricing, strengthening the business case for rollout.

However, outcomes will depend on plant‑by‑plant factors: furnace type, fuel mix, product mix and integration with existing control systems all influence achievable savings. Successful commercialisation will require validated, transparent models, demonstrated ROI in live production and partnerships with furnace OEMs and operators for integration.

Looking ahead
Deep.Meta is one of ten finalists for the Manchester Prize final in March 2026; the award and accompanying validation at Spartan UK will be watched closely by heavy industry. According to the company, its long‑term ambition is to prevent ten megatonnes of CO2 entering the atmosphere by 2030, a target that would require broad replication across multiple plants and regions.

For industrial decarbonisation professionals, the development underscores two practical takeaways: physics‑anchored AI can materially improve process control where thermal management dominates cost and emissions, and credible scaling depends on clear, explainable performance data combined with pragmatic integration pathways into existing plant control and commercial models.