How AI is reshaping the way we build, run and secure the grid

Let's Talk Energy — Episode 44 How AI is reshaping the way we build, run and secure the grid, with Microsoft’s Per Christian Honningsvåg Wednesday, 1 July 2026 SPEAKERS NB Noah Brenner — Host, Let's Talk Energy PCH Per Christian Honningsvaag — General Manager & EMEA Business Leader, Energy and Resources Industry, Microsoft [00:00] NB This is Let's Talk Energy, your go-to podcast for smart energy insights. I'm Noah Brenner. Regulators around the globe are sounding the alarm about the impact of rapidly proliferating AI-driven data centers on the stability of the electricity grid. But the same compute that's straining our power system could also be used to help it function more efficiently and access diverse sources of power more easily. That efficiency could require utilities to turn over increasingly complex and critical decisions to AI, rather than relying solely on human judgment, potentially opening up questions about safety and cybersecurity. To help us understand how AI might be used to future-proof the grid and what bumps we could encounter along the way, I'm joined today from Oslo by Per Christian Honningsvaag, who leads the energy industry practice across Europe, the Middle East, and Africa for Microsoft. We'll touch on whether AI could alleviate the need for more physical grid infrastructure and defray some of the expenses needed to make that happen. He'll unpack how a growing role for AI and grid management could improve reliability and how we should think about threats from digital hiccups or malicious cyber attacks on our critical infrastructure. And finally, he'll fill us in on how Microsoft is looking at the potential for its own data centers to at least support, if not enhance, the way the grid functions today. Per Christian, welcome to the program. PCH Thanks so much. It's great to be here. Thanks for having me. NB Well, let's talk energy. Before we dig into the details of AI and grids, could you just give us a quick idea of your role at Microsoft and how you got there? PCH There are very few places more exciting and more consequential, I would say, than the intersection of technology and energy, where this rapid technology innovation meets an increasingly complex and demanding energy system. But for me, it started in a very different place. As a teenager working at the gas station, I didn't think of it as an energy industry experience at that time, but it was. I saw firsthand how geopolitics shows up as a price change overnight, how customers react to every cent at the pump, and how essential energy is to everyday life. So to me, it's a reminder that behind all the complexity, this industry is fundamentally about serving people. And then later in my career, when I moved into technology, I spent years working with the power and utility sector in the Nordics. I still remember the day a 26-wheeler truck showed up to swap a substation because the utility had changed its view on where capacity was needed across the grid. And they were basically delaying investments by moving equipment around. And that decision wasn't made on intuition. It was basically data driven. So insights from a big data project we at Microsoft were running then with a forward-thinking partner, basically translating data patterns into real infrastructure decisions. And to be honest, Noah, that was really when it truly clicked for me because I was fascinated about the power of technology applied to energy. And 20 years on, that intersection is only becoming more critical and more exciting. So today, I get to work closely with energy companies in many countries as I lead Microsoft's energy and resources industry team across Europe, Middle East and Africa. And my team is out there every day to help energy companies understand the value of artificial intelligence and really to thrive in that intersection of AI and their energy business. NB It's really interesting. A teenager today working at a petrol station probably has a very different experience in Norway. People are probably as concerned or more concerned about the price of power as they charge their EVs as much as that price of petrol. [04:00] NB But let's go ahead and dive into the discussion. So much of the rhetoric around AI and grids is oftentimes pretty doom and gloom. But I want to talk about how AI can be used and interact with the grid itself. So let's break this down into maybe demand and supply and then also kind of the nuts and bolts of the grid. And so let's start with the demand side. Can AI be used to — and how can AI be used to — actually shape consumer demand and also to shape the demand of large users out there? And then how does that augment the function? PCH There is a lot of anxiety about AI's power use and the impact on the grid. But I think the untold story is that AI actually can strengthen our existing grids. We do see huge potential to optimize today's grid operations so they are more flexible and efficient. The power grid is evolving into one of the most data-intensive real-time systems globally. And utilities today, they do collect the data from smart meters, grid sensors, SCADA systems, grid-connected DERs — or distributed energy resources — like solar batteries, EVs, as you mentioned, but also weather and market signals. And AI can continuously learn from these vast amounts of data, predicting demand spikes or equipment issues earlier, and then orchestrate faster, more precise control actions than any human or traditional software could alone. So eventually this means fewer unplanned outages and smarter use of the power lines that we already have today. Let me bring you one good example — GridIQ. That is a real-time intelligence platform developed by the Australian market operator in partnership with Microsoft for the Australian national electricity and gas networks. So basically it serves as a unified grid intelligence layer that constantly monitors live grid telemetry and market data and alarms, providing immediate situational awareness and decision support to the market operator's control room staff. And it basically augments that human operator with AI-generated forecasts, alerts, prioritization, recommendations. And basically, the platform unifies previously siloed layers of grid operations across physical assets, telemetry and market outcomes into one model so that both AI agents and the users share a common understanding of the grid. And when it comes to that demand-side management that you're mentioning, to run the power grid smarter today includes taking advantage of that digitally connected demand side. And AI can really learn from everyday usage of consumers and work with smart devices like appliances in houses or EVs to smooth out peaks in demand without sacrificing comfort. And when you can aggregate these smaller loads from a large number of consumers, the controllable load starts to become meaningful for the grid. But to orchestrate a large portfolio of dispatchable load — where you want to take into account things like price signals, effect thresholds, weather data, grid needs, but also individual consumer preferences — this is where AI gives you that oversight and real-time intelligence that is needed to run this at scale. You're also mentioning the larger consumers. So to tap into the load of more power-intensive commercial and industrial customers is something that has been done more manually for years. And that highly depends on both their business model and to what extent the factory or industrial organization is able to shift their processes. But I think the potential is there, and fully activating that demand side has a technical potential, I think, into the several hundreds of gigawatts of capacity. [08:00] NB Does that same thinking — can it apply to these data centers that we are concerned are taking a growing share of power and can be difficult to manage? Is there a potential for AI to solve some of the pressures that AI is in some ways creating? PCH Yeah, I think when it comes to the demand-side management of data centers, we are definitely exploring it and we are exploring how data centers might provide services to the grid in extreme conditions, as long as we can guarantee no impact to the customer workloads that are actually running in the data center — because that is also the workloads of energy companies. So Microsoft's long-term vision here is to leverage data center assets in ways that enhance grid reliability and sustainability in partnership with utilities and regulators. You might have seen some of our projects on grid-interactive batteries as one example. When needed, we can quickly transition to backup power sources to reduce impact on the grid without impacting customer workloads. Or in some cases, we have offered grid-interactive backup batteries to help grid frequency as well. But our first priority is always maintaining customer trust and uptime so that any steps towards demand-side management are optional and carefully controlled. We collaborate with grid operators to align on solutions that work for everyone, solutions that establish long-term confidence in newer technologies. And we see this as an opportunity to support decarbonization efforts while keeping our reliability promise. NB I want to shift over a little bit to the supply side. You mentioned these distributed energy resources — DERs. This is solar. It might be solar panels on a residential home or on a commercial building. Could be wind assets. One of the things that struck me as we were talking, preparing for this episode, was this idea that these distributed energy resources become revenue generating, not simply avoiding costs, which was a line that you had, I think, in a presentation. How does AI enable us to run this sort of network of supply more efficiently? PCH I sit on the advisory board with the Copenhagen New Infrastructure Partners and a couple of months ago we published a study on end-user flexibility. That study was based on households in Denmark. An average household in Denmark with rooftop solar and at least one EV can annually save 11,000 Danish krone by just shifting existing loads based on pricing signals. And they can also earn up to 18,000 Danish krone — meaning more — by selling their excess power back into the grid. So there's clearly an opportunity to participate bidirectionally for prosumers. We also work with another partner called NODESmarket, who has created a flexibility trading platform for local flexibility markets. And flexibility service providers — either large individual companies or consumers through a so-called aggregator — can participate at all levels in the grid. And in the local markets, they have established DERs as revenue generators for the suppliers, offering into the market where grid companies place real-time bids. So AI has taken this key role in these trading scenarios, especially at volume and scale. NB What about simply moving power around more efficiently? Are we seeing AI finding ways to get power from point A to point B in a way that's better than maybe we've done traditionally? PCH There are definitely some clear advances in that space nowadays as well. Actually, Microsoft Research — our research arm within our company — launched last couple of months something they call GridSFM. So that is a lightweight AI model developed by Microsoft Research that acts like a digital simulation engine for the power grid. So it can predict the optimal electricity flow at any point across large transmission networks within milliseconds, enabling utilities and grid operators to test far more scenarios, reduce congestion, and integrate more renewables into the equation — and make grid planning and operations significantly more efficient. And this AI model is trained on hundreds of grid topologies [12:00] PCH and over 500,000 operating scenarios. The way to think about it is: think of how ChatGPT learned language patterns instead of looking up every sentence in the rule book. So GridSFM then gives grid operators and planners the ability to evaluate many, many more possible solutions and ways for the power flow much faster, and then eventually leading to better operating decisions. NB If you can put this in a bit of historical context for us and draw on some of your experience — grids have benefited from digitalization for years and we've used the terms "digital" and "AI" so far in the conversation. But I feel like AI and digital are oftentimes maybe used interchangeably when that's not necessarily the case. And so what is it that AI specifically does that couldn't be achieved just with a more simple or more traditional digital solution? PCH It is true. Digitalization of the grid has been underway for decades. And we have had SCADA systems, smart meters, automated switches. Those are essentially rule-based or remote-control technologies that gave us a lot of efficiency. But you can look at AI as the next step in that evolution, if you will. It goes beyond pre-programmed logic. So AI can learn from data, it can spot complex patterns and even generate new solutions dynamically, which classical software might miss. For instance, we tried for years to speed up energy project permitting through traditional software and frankly hit the wall. But when we applied generative AI to the problem, it could quickly read thousands of pages of regulatory text and draft permit documents within minutes — something that used to take experts months. So that agile information-synthesizing ability is unique to AI and wasn't possible with simple digitization. Nuclear, as one of the power sectors, has adopted AI at scale. And the only thing that might be more complex than building a nuclear plant is designing and permitting one. So permitting alone can take years, cost hundreds of millions of dollars and involve an immense amount of data processing and reporting. And now we have nuclear companies that have reduced the time-intensive permitting process by 92% using our generative AI for permitting, saving between $80 and $100 million per year. So if permitting of nuclear can be accelerated by AI, I think we can achieve similar benefits applying AI to permitting processes within both renewables and grid connections as well. And recently we have, together with some of the leading global grid companies, performed simulations where AI was given the actual alarm and telemetry data recorded during some of the historical large power station failures which caused widespread blackouts. So meaning — take new technology, look at what we had available of data and telemetry in the past. During that simulation, AI surfaced critical details in real time, including signals that operators recalled missing during the original incident. So AI automatically produced clean structured timelines broken down by power system. It generated diagnostic theories and contextualization using the operator training materials. And then the end-to-end analysis was produced in approximately 15 seconds. So the result exceeded all expectations and validated that this integrated AI approach could deliver outcomes and intelligence not previously considered feasible in live grid operations. NB Can it help us avoid needing to develop infrastructure? We hear these astronomical sums that are needed to modernize the grid. Are there ways that we can use the software to replace the need for new hardware? PCH There are also limits to it. Let's be clear on that — because we will always need to invest in physical infrastructure for the energy transition. [16:00] PCH Digital solutions can definitely delay or reduce significant amounts of that hardware growth, buying us time and also saving money. So I don't think it's an either/or — it's an "and both." No amount of software will give you a new transmission line or a power station if you just don't have enough capacity built. But what smart software and AI can do is squeeze every bit of efficiency out of existing assets so that we can serve more demand with the same equipment. And it can also pinpoint where new investments are most needed, avoiding overbuilding in one place while ignoring another. One example is dynamic line rating — and that is a technology used to monitor the real-time capacity of power transmission lines. So DLR technology provides a more accurate real-time assessment of the transmission line capacity, enabling the full use of the line without compromising operational safety. So this can help utilities optimize the use of existing infrastructure and defer costly upgrades. But it also reduces the risk of overloading the line, which can lead to blackouts and other disruptions. There are companies we work with — like Heimdall Power — that are already doing DLR projects in multiple countries. So this is not a new, unproven technology or solution space, but now the data that sensors are transmitting to the cloud is interpreted with new AI capabilities that really makes these solutions even more precise and more robust for real-time operational tasks. NB I'm wondering — we've talked about where AI does work well, where we have seen the gains, where we're likely to see them in the future, but where doesn't it work well? Or where are some places that maybe AI hasn't advanced to the point it needed to yet, or maybe it just simply isn't going to be the right approach in certain parts of the grid? PCH We are focused on helping utilities establish governed data foundations that support both analytics and AI across operations, field work, even customer engagements. That includes unifying both IT and OT data as well. So in many organizations, this data remains distributed across systems with inconsistency in definitions and varying latency and even uneven governance. Without a consistent and trusted data foundation, AI initiatives basically struggle to scale beyond isolated use cases. We see tasks where conventional digital solutions work fine. So we don't force AI for its own sake. There are enough opportunities to apply AI where it directly improves reliability, affordability, and productivity, and doing so with the security and governance necessary for critical infrastructure. The other point, when we talk about whether AI works well, is to be clear on what AI we use. So the large language models that we all know through chatbots and so forth — they work really well for reading, for summarizing, for exploring information. This is because they are trained on books, websites, documents, and so forth. To put these into an industrial environment comes with a risk of them working without context. So this is where the domain-specific foundational AI models come in. These are models that you train on your data and your structures to enable them to predict and optimize and simulate. We have launched [20:00] PCH two of those grid-relevant open-source domain models this year. One is Aurora, which is a foundational model for global weather and atmospheric forecasting. So in addition to being able to predict extreme weather events — which is useful for both operations but also for long-term planning of infrastructure — this model is way faster than traditional models. So it brings that opportunity to be impactful in real-time power production and real-time grid operations as well. And the other open-source domain model that I mentioned — GridSFM — which can predict optimal electricity flows across large transmission networks in milliseconds. NB We've referred to the grid multiple times in our conversation as critical infrastructure. And I know people are concerned about the reliability and the security of using AI to manage critical infrastructure. So talk to me a little bit about how human oversight works in an AI-optimized grid and how you think about the need for human judgment — having a human in the loop, so to speak — as AI advances. Are we going to get more comfortable turning over more processes to AI? And where do we need to have a human touch there? PCH Trust and oversight is absolutely central to how we deploy AI in the grid or in critical infrastructure as such. So we are not about to let a black-box AI just run off and control critical equipment without guardrails. Human experts remain in charge and we keep that human in the loop — safeguards by design. We use AI to augment their decision making. For example, AI might crawl through thousands of sensor signals and then suggest an action plan, but a grid operator will review it and approve it. So we design these systems to explain their recommendations and back them up with data. Over time, as AI proves itself and our comfort grows, more routine decisions might be delegated to these smart systems. But I foresee that human oversight will always be part of the equation for critical infrastructure. Even though we trust autopilots, even when we fly planes, we still keep a trained pilot in the cockpit. The bottom line here is that we manage AI in the grid with a very high bar of safety, transparency and human control. NB Are there regulations around how much of the grid's functions and processes can be automated through AI? Should there be? PCH Utilities are a heavily regulated sector. I think the regulatory landscape for AI in grids and grid operations is still evolving. At present, there aren't many AI-specific rules that say no more than X percent of the grid can be run by AI. But what we do have are very strict regulations around reliability, cybersecurity and safety. And those effectively impose limits on automation. We want to encourage innovation, but with guardrails. So in practice, this might mean certification of an AI system for grid control, or requiring humans to always have ultimate override. I think the key is to balance safety with innovation. So we need AI's help to manage the complex, dynamic, modern grid. But it must be done in a transparent, regulated way so that everyone — regulators, utilities, customers — has confidence in the outcomes. [24:00] NB You mentioned cybersecurity there as a key security risk, and certainly the worries about cybersecurity and the grid have grown as the world quite honestly becomes a more contentious place in many ways. Is an AI-integrated grid more or less secure than a digitalized one, to outside attacks? And if you could address both the ability to repel a cyber attack as well as to maybe repair or get back online if the grid does experience one. PCH Security is paramount for any grid technology, AI included. So on one hand, a more digital, AI-driven grid does introduce new cyber considerations. If you have more connected sensors, more software in control loops — yes, the attack surface can increase. So that's why we design with a security-first mindset. We partner with specialists like Dragos on the operational technology side, and apply zero-trust architectures to every layer. In other words, we assume breach and we constantly verify. On the other hand, AI can actually make the grid even more secure. So AI can monitor network traffic and grid behavior in real time, flag anomalies and intrusions far faster than any human might. Microsoft's own Security Copilot, as an example, has been shown to reduce incident response times by 30% for companies. In the grid context, that could mean detecting a cyber attack early and then isolating affected equipment automatically before it spreads. So while an AI-integrated grid must be heavily secured like any other digital system, it also becomes an active defense tool. So it's never either/or. We do incorporate robust cybersecurity measures and AI-driven defense so that as the grid gets smarter, it gets safer too. And as for self-healing or getting back online, AI can play a role in that too. In a physical sense, we already have self-healing grid schemes where, if one line goes down due to a fault, power reroutes itself. And AI can enhance this by recommending optimal restoration steps and even managing predictive maintenance to fix vulnerabilities before an attack happens. NB I want to shift gears a little bit here and talk about data centers and their own interaction with the grid. You mentioned at the top of the program that there's potential for data centers to be grid-stabilizing, to be a grid asset — not necessarily a grid liability or a major demand. So what does that look like? What does a grid-friendly data center perhaps look like in terms of its physical infrastructure as well as its function? PCH Data centers have traditionally been thought of as big energy users, but we are seeing many examples of the same data centers giving back to the grid. Batteries as part of the UPS for data centers, for example, have been enabled to do real-time interactions with the power grid. This can reduce the need for traditional natural gas-fired power plants to maintain spinning reserves that can quickly respond to provide grid services. And in an evolving energy system where intermittent renewables result in a more volatile grid frequency, the value of grid-connected batteries increases. Another example is how waste heat from data centers can be connected to the district heating system. So in Finland, Microsoft is now building our new data center region there, where waste heat will cover approximately 40% of the heat demand of 250,000 customers in the area around the data center. So this helps take some of the load away from the power grid as electricity-to-heat consumption is then reduced. [28:00] NB Microsoft unveiled a five-point pledge earlier this year on how data centers can coexist with communities. And the first point deals with power and the grid. It says — exactly — "We'll pay our way to ensure data centers don't increase your electricity prices." And there are a number of provisions underneath that. But I thought one was particularly interesting. It said: when our data center expansion requires improvements in transmission and substation capabilities, we will continue our existing practices by paying for these improvements. Are we going to see Microsoft and other hyperscalers play a bigger role in grid modernization going forward, to enable these massive investments that are being made in computing power? And what does that look like in practice? PCH I do expect all hyperscalers like ourselves will collaborate more deeply with the energy sector on many different levels. We are already seeing tech companies hiring energy experts and vice versa. And this community-first AI infrastructure initiative that was launched last year in response to concerns that AI data centers could impact local communities through higher electricity demand, water consumption and land use — as part of this initiative, we are committed to collaborate with local utilities, provide early transparency around our projected power requirements. We want to ensure that infrastructure development strengthens local systems rather than adding pressure to them. So when our data center expansion requires improvements in areas like transmission and substation capabilities, we will continue that practice by paying for these improvements. But even beyond funding grid upgrades, we are also contributing technology and expertise — for example, providing cloud platforms and AI tools to utilities directly, but also to the suppliers of operating software, like partners like GE Vernova and Schneider Electric, to accelerate grid digitization and optimization. So yes, you will see Microsoft and our peers at the table with regulators and grid operators, co-innovating solutions for better energy forecasting, planning and reliability. After all, we can't support these advanced AI services if the grid isn't up for the task. So it's in everyone's interest for us to play a significant, proactive role in modernizing it. NB One of the biggest discussions around AI and power is essentially power that doesn't interact with the grid at all — so-called behind-the-meter power. I'm wondering how you think about the balance of behind-the-meter versus grid-connected power for your own facilities at Microsoft. Is putting power behind the meter a strategic choice for data center developers in general, or is it more of a coping mechanism when we fail to kind of get the grid right? PCH At Microsoft, we have been in favor of being an integrated part of the grid and the grid ecosystem and not operating in isolation unless it's absolutely necessary. On-site generation or behind-the-meter generation can be useful in certain situations. We see it mostly as a supplement, not a main solution. The ideal scenario is to work with utilities and governments to ensure that the public grid can be the one delivering all the clean power that we need — because a robust and well-planned grid benefits everyone and not just one facility. So our strategy has been to invest in grid-connected solutions, and you also see this [32:00] PCH proven in the enormous amounts of renewable energy PPAs that we have been signing in the market to add utility-scale wind and solar to the grid. NB Per Christian, it's been a fascinating conversation. I want to start to wrap it up here. There was something Microsoft's founder Bill Gates said in June of 2024 that really struck me. He said it in London during climate week, if I remember right. He said: "Ultimately, AI would save more power than it consumes." And he cited many of the reasons we're talking about today — better, more efficient power generation, transmission, and usage. It's been almost two years to the day since he said that. Do you think that statement is still possible? What have you seen over the past couple of years that says that could be the future we're moving towards? Or conversely — in 20 years, are we going to look back and say that AI has solved some of the problems that we were worried it would create, or will we see it as kind of the start of some of the problems that we're worried it might create when it comes to power and the grid? PCH Yeah, I do share Bill's optimism, first of all. His point was essentially that, yes, AI uses energy, but it also drives bigger efficiency gains elsewhere — enough to more than balance out its own footprint. Along the lines that we have already talked about: AI helping grids avoid waste — like reducing renewable curtailment, congestion costs, speeding up permitting to deploy clean energy faster, optimizing power plants to burn less fuel, enabling more demand response, and so on. All of these gains add up. Meanwhile, data center operators, including ourselves, are pushing hard on efficiency — like developing new chips and new cooling technologies — and even in software development. So AI helps to build software code that is more efficient to run, so the hardware consumes less energy while running this new AI-based software. But the impact of AI extends well beyond energy and data centers as well. Think of transportation and supply chains — AI can dramatically improve route optimization, drive significant efficiency gains, like reduced energy consumption. In our latest sustainability report, you can read that our scope one and two emissions are down by 30% from the 2020 baseline. And this is based on [36:00] PCH our work on clean energy procurement and the collaboration with the energy industry to add new renewable energy onto the grid. But you can also read about our scope three, which has increased by 26%. And that is based on the rapid expansion of the infrastructure and the buildings and all that related to data centers. So key sectors to continue to invest in decarbonization and collaborate with are, for example, steel and concrete. So our scope three — which excludes our power use — represents more than 97% of our carbon footprint. So there's a huge potential to work on reduction initiatives, even beyond power. To build on Bill's optimism: some of the greatest advances that we have seen AI contribute to so far, you will find within material science — material science that can help the energy industry on innovation related to things like battery technologies or new carbon capture techniques. But it can also help value chain companies on delivering new green products like steel and cement that at the end of the day will show up as a reduced carbon footprint. NB Is there anything that we've missed? Or is there a parting thought that you'd like to leave our listeners with around this topic? PCH AI is not just reshaping technology. It is reshaping how we are generating, managing and consuming energy as well. And that's why the future of AI and the future of energy are increasingly the same story. NB Per Christian, thank you so much for joining us today. PCH Thank you. [40:00] NB Thanks for listening to Let's Talk Energy. This podcast is a Rystad Energy production produced by Elliot Busby and Bade Og. Check out the show notes for further analysis on the topics we've discussed in the episode and find us on social media — we're at Rystad Energy on all your major platforms. While you're there, leave us a review, click that like button and subscribe. You can also keep up to date on our website. If you'd like to send us questions or maybe you have an idea for our next episode, go ahead and email us directly. It's podcast@rystadenergy.com. And just a heads up, we'll be taking a short summer break and adjusting our schedule for the remainder of 2026, but we'll share more on that soon. As always, don't forget to join us next time. For more, Let's Talk Energy.