AI-Powered Hydropower: Enhancing Flexibility, Efficiency and Value
Insights
- Hydropower is the single largest source of renewable electricity in the world, more than wind, solar, and bioenergy combined, yet it remains among the least digitized parts of the energy system.
- The hardest constraint on predictive AI in hydropower is the absence of failure data, because operators run conservatively and rarely break equipment, leaving no training examples at the exact point where models would create the most value.
- Many hydropower use cases are not unique to the sector, and the computer vision and predictive maintenance pipelines built for oil and gas transfer directly, which leaves reservoir inflow and water valuation as the genuinely specialized problems.
Recorded at ONS 2026 in Stavanger, Norway, this panel discussion features Åsne Taksrud, Head of Technology Market at Å Energi, Ole-Petter Bjørnstad, Operations Manager and Technological Developer at Tinfos AS, and Jon Olav Åbeland, Senior Technical Program Manager at Microsoft, in conversation with Joseph Alenchery, SVP and Business Head of EnergyNext at Infosys. The discussion addresses a problem specific to hydropower: operators want AI to tell them how hard they can run their machines, but decades of cautious operation mean they have almost no data on what failure looks like. Taksrud describes the goal of a full digital twin covering the plant, the rivers, and the market, while explaining why every plant is effectively bespoke. Bjørnstad puts the stakes in operational terms, noting that large plants are tailor-made, spare parts often do not exist, and a single machine failure can remove a unit from service for a year or more. Åbeland brings the view from oil and gas, arguing that the OT to IT connection was the real barrier and that the patterns now exist, leaving execution as the open question. Together the panel examines where generic AI capability ends and sector-specific modeling begins, and what governance and risk control it takes to scale in critical infrastructure.
Joseph Alenchery:
Enhancing efficiency, flexibility, and value.
It's a tongue twister of a topic, but at the root of it you have hydropower and you have AI and the marriage of these two forces coming together.
My name is Joseph Alenchery. I head energy as a segment for Infosys. But nothing excites me today than this topic of hydropower.
In fact, hydropower is a single largest source of renewable electricity in the world, more than wind, solar, and bioenergy combined. And that's a big, big statement to make.
China, 30% of world's hydropower, around more than a thousand terawatts is being generated in China.
And one of the biggest producers of hydropower after China is of course Brazil, you know, which is again big time into hydropower.
Now we speak from this country, Norway. 90% of electricity in Norway, grid electricity is generated from hydropower. So it's a topic of great relevance.
AI in hydropower, if when you imagine an AI powered future for this industry, Åsne, what is your take?
Åsne Taksrud:
We are already doing a lot of predictive maintenance and predictive hydro, predictive weather forecast and the pricing and what we should do when.
But at the same time there's always sort of a man in the middle, or know-how, at the end of the loop. So I think going forward we will be more GenAI. Like more agents, more automated processes.
And I would say the dream scenario for Å Energi is to have a fully digital twin of the hydropower plant environment, with the rivers, with the market, with the players, all together in one sort of digital model.
Joseph Alenchery:
Ole-Petter in your experience, do you think AI has a role to play in accelerating this whole industrial progress?
Ole-Petter Bjørnstad:
Yeah. We use a lot of machine learning in modules, and that we have done for 30 years at least.
We use AI to get all the historical values from the machine, all the data from the weather and everything, we put that into a module. And we calculate how that can be more and more efficient.
Joseph Alenchery:
And Jon Olav, what are you seeing in terms of AI in hydropower?
Jon Olav Åbeland:
There's a lot of use cases that are relatively generic that I recognize from oil and gas, right?
If some of the use cases, like in hydro, they have cameras, they're looking for inflow problems like branches blocking, or they're looking at fish going by, invasive species, all this kind of camera stuff.
The pipeline you use to analyze that is the same pipeline that you would use in Aker BP to analyze film from a robot dog going on a platform, right?
So there's a lot of generic things, like the predictive maintenance that was mentioned here, that's nothing unique.
So we bring that experience into hydro as well.
And then you have the more specialized use cases like how much of the rain will actually reach my reservoir and when, and what is the value of this water, and how should I plan the operations to spend that water?
Joseph Alenchery:
You spoke about a very important vector, that is we have years of experience, decades of experience using AI or emerging technologies in oil and gas, and to be honest, many of that's applicable in the context of hydropower.
And of course, there are nuances. You also spoke very importantly about partnerships. This industry has been there for 100 years. You know, hydropower and to varying degrees we have applied technology in that.
But today we're at the cusp of what all of you are describing as kind of an AI-powered future.
What's the big challenge you're seeing in this context? I mean, what's that one thing? Is it old systems, old machines, old sensors? No data? What's the big thing in front of you, Åsne?
Åsne Taksrud:
Probably a bit of everything. But I think one major challenge for hydropower plants is that they are sort of unique.
And another thing is, of course, that when we are going into AI and more predictiveness, the one thing that we want to sort of use this for is to how hard can we drive without breaking something. So this is our tipping point.
But we have very little data of when something breaks, because we are not breaking anything. We're always sort of driving it on the safe side.
So we don't have training data on the things that we don't want to happen. So we have to model something that we can't train for. So that's a challenge.
Joseph Alenchery:
So Jon Olav, what is your perspective on this?
Jon Olav Åbeland:
I think the thing that held us back in the start is this OT/IT connection is difficult. This is critical infrastructure. This has to be 100% one way street where the sensor is sending data to cloud. There is nothing going the other way.
So we have patterns with this. We work with this. But this was a big step to get past for customers and there are still some customers working on this.
But I feel, we have the patents now and now it's a question about doing it, and doing it right. There are no shortcuts in this area.
Joseph Alenchery:
So Ole-Petter, that's the last question I'm going to formally ask you before we conclude. What challenges are there in your eyes?
Ole-Petter Bjørnstad:
The challenge is, as you said, Åsne, is that if you break a machine, it's dead for one year or more, and it's extremely costly.
So everything we do must be risk assessed, a lot of times, because a failure is fatal, because all large hydropower plants are tailor-made.
And we don't have spare parts. We don't, on some, we don't even have bearings, because they're so large. So failure is fatal, in fact.
Joseph Alenchery:
We have to push forward, the world doesn't stop and wait for us, and if we don't do it, some competitor will do it, right?
So we just have to make sure we do everything with control, security, compliance, governance. All those things need to enable us to scale, in not only volume, but in complexity and projects as well.
Fantastic. And Ole-Petter?
Jon Olav Åbeland:
Yeah. And asked all industry today hydropower has to do this, because this is the only solution to get more kilowatts out of the hydropower plants to use them smarter.
We can't just use them as we did in the old days. We need AI, we need machine learning. We need smart modules to attack the markets, to attack the flexibility.
It's needed. So what we do is the future.
