Industrial AI, Hype or Reality?

Industrial AI means something different depending on who's using the term, and this piece breaks down what's actually behind it.

Attila Dávid
Attila Dávid

Director of Sales Engineering at Akselos

industrial-AI-hype-or-reality

The man who founded Uber has decided the next big thing is not an app. It is steel, concrete and machinery. Travis Kalanick now argues that mining, construction, heavy transport and food production are the industries waiting to be transformed, and that industrial AI is, in his words, getting ready to power the next industrial revolution.

Travis is not alone. Siemens has built an entire event brand around the phrase. Hannover Messe ran its 2026 edition on the theme. South Korea launched national industrial AI projects in June 2026. Aramco uses the term about itself. In one month this summer, industrial AI appeared in more than 2,500 online articles.

So it is worth asking what people actually mean by it.

What is Industrial AI?

Industrial AI is artificial intelligence applied to physical operations rather than digital ones. Not writing emails or summarizing documents, but running plants, predicting failures, optimizing yield and deciding when a piece of equipment needs attention.

The difference is consequence. When a chatbot is wrong you get an awkward sentence. When an industrial system is wrong you get a shutdown, a write-off or a safety incident. That single fact shapes everything about how these systems have to be built, and it is why almost none of them are pure machine learning.

How companies are using it

Siemens uses industrial AI to mean scale and connection. It attaches the phrase to CNC controls, engineering agents and shipbuilding partnerships, on top of an installed base it puts at more than 35 million connected devices. The proposition is one intelligence layer across hardware, software and connectivity rather than a shelf of separate products.

AspenTech uses it to mean better models. Aspen Hybrid Models blend live plant data with machine learning and first principles simulation. Aramco deployed them across its refinery planning in April and reported yield and quality prediction accuracy of up to 98.5 percent in key units. Emerson followed in May with AVA, an agentic platform built on the same domain models.

Kalanick uses it to mean robots. His version is software, sensors, robotics and AI automating the operations of entire sectors, which is a story about replacing physical labor rather than understanding physical assets.

Comparison of how Siemens, AspenTech, Kalanick and Akselos define industrial AI
Comparison of how Siemens, AspenTech, Kalanick and Akselos define industrial AI

Akselos and Industrial AI

Where Akselos fits is easiest to see in a single piece of equipment.

Take a steam methane reformer. Hundreds of catalyst-filled tubes operate in the creep regime, where steel slowly deforms over time. Even a small temperature change has a significant impact on creep life.

No one can see the true temperature of every tube at every moment, so engineers take the conservative route and assume the worst case for all of them. Good tubes get replaced early, the few that really are running hot are hidden in the average, and the unit runs below what it could do.

akselos-smr-model-simulation

Akselos’s industrial AI starts by establishing what the steel is actually going through, and every asset needs a different way of seeing:

  • Reformer tubes: a heat balance built from inlet and outlet temperatures and the known heat sources and outputs gives each tube’s temperature, calibrated against infrared camera measurements.
  • Thermal mix points: CFD.
  • Coke drums: machine learning trained on historical laser scans to predict how bulges will grow.
  • Behind all of it: decades of operating data from process historians.

None of these methods is new on its own. The expertise is in picking the right ones and combining them so the model behaves like the real asset. That’s the closest picture of reality you can get without cutting the steel open.

Physics then turns those conditions into stress, creep and fatigue damage, and remaining life. In the reformer, the creep model is calibrated against measured tube dilation, so the predicted damage matches how the tubes have actually grown.

This is why the approach has to be hybrid. Machine learning only knows the past it learned from, while physics can describe conditions the asset hasn’t seen yet.

But conventional finite element analysis takes hours per case, far too slow to run through years of plant history. Akselos’s reduced-basis solver, RB-FEA, developed from research at MIT, solves the same physics almost instantly and with quantified accuracy.

AI then helps engineers question the results: which tubes are using up their life fastest? What led to the drum’s worst fatigue cycle?

Back to the reformer. Instead of assuming the worst case, you know how each tube is actually ageing.

Is there room to push the unit harder? Or how far do you need to dial it back to reach the next turnaround?

Akselos Structural Performance Management (SPM) gives engineers the insight to answer those questions with confidence, grounded in physics.

Attila Dávid
Attila Dávid
Director of Sales Engineering at Akselos

Attila works with offshore operators and engineering teams to turn structural insight into practical asset decisions. He focuses on helping customers improve integrity strategy, remaining life assessment, and operational confidence across complex O&G assets.