Turning a Yield-Co into an AI-Native, Next-Gen IPP with David Willemsen of Tion

Speaker A: Foreign.

Speaker B: Welcome to the pexpark Podcast. Every two weeks we'll be bringing you fresh updates and insights on the renewable energy market, along with a guest who'll share unique perspectives on the critical trends shaping our industry.

Speaker C: Welcome to a new episode of the Paxa Park Podcast. For years the pitch for owning renewables was buy the de risked assets, manage them lightly, distribute cash. That model is now on the visible strain, with many listed ipps and yieldcos trading below their net asset value. And so the model is changing. While buying the assets remains a core function, what's changed is what it means owning them and where the value is generated. Joining me today is David Willemson, Chief Commercial Officer at Theon Renewables, a uh, European IPP that develops, builds and operates onshore wind, solar and battery storage and that has spent the last three years rebuilding itself from a yieldco into something quite different and exciting. We get into what a revenue organization actually looks like when you build one from scratch. Why? David argues asset management isn't a support function for the investment thesis, but is the investment thesis and what he means by a, uh, digital ipp, including which software categories he is now building in house instead of buying, and where he thinks the limits of that are. His story of digitizing an IPP is worth the episode on its own. And as I do every two weeks in the second part, I'll take a look at some of the excellent reporting from the PAC support team. This time we're diving into queue reforms in PJM that could usher in a huge storage boom. And we look at the impact the European CBAM regulation is having on um, power prices and PPA activity in Southeastern Europe. Today.

Speaker A: I'm very happy to have with us again a Paxo park alumni, David Williamson, who embarked on a very interesting career after Pexo park. And I'm very glad to have you with us.

Speaker C: Welcome David.

Speaker A: Uh, to be here, David, before we

Speaker C: dive in, who are you?

Speaker A: What did you do after Pexapark?

Speaker C: Who is Tion? What's the company up to these days?

Speaker D: Yeah, after Pixar Park 24, I joined uh, Tion Renewables as Chief Commercial and now also Chief Digitization Officer. My job sits at the inter intersection of the commercial strategy, energy management and increasingly the digital infrastructure that connects the two. Theion is a European energy producer. We develop, build and operate onshore wind, solar and battery storage assets across Europe together with the portfolio of Clearweiss, another European IPP whose operations we manage. Since the beginning of the year, we manage around 900 megawatts of operational Projects. Our largest markets are uh, Germany, Spain, France and Poland. And we started our life as a yield core. So the idea was lean management de risk, cash flows, infrastructure like returns. But that model has evolved significantly since eqt, one of the world's largest private equity companies in the infrastructure segment took us over in 2023 and we've made a full shift to like a proper IPP model. It's been a three year journey that fundamentally changed how we think about operations, data and value creation.

Speaker A: Right. And this we're going to talk about. So you already started a bit with uh, a sector view. We have seen quite a lot of those yield cos and other publicly traded ipps trading way below net asset value. Open story in the market, a lot of stress. What's your read on the story and now why has this been happening?

Speaker D: I think it reflects the reality that the market has slowly been pricing in the development and acquisition phase is not the hard part. The hard part is what comes after. So a lot of IPPs and funds deployed capital heavily, especially in the years between 2020 and 2023 at the height of the market. The with high valuations, sometimes aggressive return assumptions. Now the interest rate environment then made refinancing more expensive. But I think the deeper issue is operational projected returns in renewables are only achieved if you actively manage them. If you treat an asset like a bond that just pays a coupon, you leave a significant amount of value on the table. And the market is now kind of sorting companies on that dimension. Those who can genuinely extract value from operational portfolios versus those who assumed the model would take care of itself.

Speaker A: That's the big difference basically having more. I mean you can't do low effort asset management anymore. There's more action to that.

Speaker D: Yeah, exactly. I think traditionally, you know, companies have run this rather on the lean side and I think now three things are happening kind of simultaneous simultaneously. The technology complexity. So assets are just no longer just solar panels feeding into flat tariffs anymore. Uh, you have batteries, hybrid structures, colocation, each with their own optimization, logic, regulation. So new rules are coming out on a weekly basis across European markets. You're dealing with redispatch ancillary service frameworks, curtailment regimes and also increasingly complex settlement mechanics. On the offtake side, the revenue streams are also evolving. Now you don't have only the day ahead market and PPAs, you have things like frequency response, inertia, capacity mechanisms, congestion management. And none of these existed at scale for kind of the renewable world five years ago. And each requires a Dedicated expertise to participate in. So if your asset management is focused on making sure the assets are running and the invoices are being paid, which is honestly what most of the industry was doing in the last decade, you're kind of not even playing the game on those additional revenue dimensions that you often need to achieve the high return expectations that you have.

Speaker A: I mean as you mentioned, most of the IPP started as a yield core, let's say as a similar construct. When you look back, do you think there was like kind of a wake up moment or wake up call or was it all incremental change? You know, a bit more merchant exposure, a bit more complexity here, a bit more batteries there or how would you. It's always easier looking back and trying to make order of the things.

Speaker D: Yeah, I don't think there was this single inflection point. It was rather happening gradually across the industry with every little change, every new additional piece of regulation or market revenue source. I uh, think after we started out we pretty quickly realized that this yield core premise was elegant. Right. So it was buying DE risk assets, manage them lightly and distribute cash. But the DE risked in renewables is increasingly a um, kind of relative concept. So markets evolve, regulation changes and the moment you need to make a commercial decision under uncertainty, you know what to do about curtailment, how to treat negative prices and so on. You need internal capability. You can outsource the process, but you can't outsource the judgment. And so what we realized was that asset management and commercial capability aren't support functions for the investment thesis. They are the investment thesis. That's the mental shift that happened at TION and that drives kind of now everything else that we do.

Speaker A: Yeah, when we were together at Paxa park and then um, when you embarked on your next step, I think a lot was focused on, on revenue management, what we call the next gen ipps, organization building, so called revenue brains. Uh, could you share a bit light on um, how you built this in practice or how it looks like today for an IPP like tion?

Speaker D: Yeah, of course. I mean our setup is still relatively lean in the sense that we don't have our own market access so we don't trade actively on the exchanges, but we use third parties to, you know, do our balancing and give us access to futures markets and so on. But at the core we separate three functions across two teams. So we have the commercial origination that is PPAs, FPAs generally offtake structures largely focusing on our new uh, investments rather long term. And then we have the energy management which covers the, let's say day to day market operations, the balancing tendering and also managing our hedging strategy across the portfolio which is a bit more short term oriented. So one to maximum, three years ahead typically. And then there is the whole data and reporting layer that ties everything together and gives the leadership near real time visibility into both the operational and financial performance. So the commercial team currently consists of four people at the moment and our IT and digitalization team of IT consists of three people. And then on the, let's say infrastructure side we run a tech stack that is heavily reliant on kind of a, um, single provider, um, because that allows for easier integration of the different tools. The centerpiece of IT is our data lakehouse which acts as a single source of truth for the entire business. Um, so production data flows in from our uh, digital twins monitoring platforms, the meters we have connected. But critically we don't just aggregate metering data, we've built the full revenue logic on top of it. So curtailment calculations, automatic uh, automated invoice reconciliation, market settlement verification, all of that flows into kind of a unified reporting environment. And the goal is that at any point in time we know the operational and financial state of every asset in at least near real time.

Speaker A: Yeah, in our prep call you mentioned this to be the digital IPP and we're going to deep dive a bit in that. Before we go there, I was just wondering on um, your revenue organization, like how you set risk and trading limits, how you look about this, is this a central separate function or just to sit with the CRO with you or how you go about that?

Speaker D: Essentially it sits within the commercial team, but of course we act within agreed, uh, let's say guardrails. So we agreed a hedging policy that of course everyone in the management signed off, but also our board signed off on so that we have the flexibility and the framework to act quickly when needed. I mean if we look at especially things like short term hedging, you have to be quick. If you go to a trader or utility and you ask for a price, the validity is often 15 minutes. Uh, so you don't want to have to call your bot and ask hey, can we, can we lock in these prices? So we, we have a, let's say dedicated framework where we define which products we use. Uh, so as produced pay, as indexed as, as nominated contracts for the most part, uh, which tenors we can cover at which prices, what kind of risks we, we can take and what kind of risks we typically want. To stay away from. And that framework works quite well. It's, it's lean and flexible, but prevents that we, you know, do anything that would not be in the interest of, of the company and our shareholders.

Speaker A: Right. You mentioned there wasn't really one inflection point in this evolution. But you had a takeover. I mean there was the AQT takeover, which probably was an inflection point for tion. Um, and how did that change how you operate? What did it enable?

Speaker D: Absolutely. I mean for us this was for sure such a critical point in our history. And I think I would summarize it that EQT brought two things that we didn't have before. Rigorous governance expectations and an um, unambiguous mandate to grow through value creation rather than just by buying new assets. On the governance side, the reporting cadence, board level scrutiny of operational KPIs and the expectation that you have to explain every variance from the plan that forces a level of data discipline that most asset managers, including us, didn't have. So you can't walk into a board meeting saying, hey, we think returns are roughly on track. With eqt, you need to show exactly why, what the drivers are, uh, and what you are doing about any potential gaps. So the pressure was honestly the forcing function for us to fully embrace automation and AI. We couldn't hire our way to the reporting quality that uh, EQT expects. So we needed to systematize it. And that turned out to be the right constraint for us at the time. Right.

Speaker A: You mentioned in this discussion that two functions moved center stage. One is the revenue organization, revenue management and then asset management. And what does this look like on the asset management side?

Speaker D: So it's essentially first in housing many of the capabilities. And for us that means the commercial asset management. We still outsource the technical asset management. But what it does is it allows us to look very critically on what do we have to do to improve the asset's performance. We don't want to just keep operating it. We, and only we have the incentive that we perform according, uh, to the expectations that the business cases set for ourselves. And I think that is in general a key differentiator. Um, it's incentive. Right. So traditional asset management and also often O and M service providers. Fundamentally it's a billing effort. Right. You have hours, headcount reporting cycles and the revenue of these companies is stable as long as the assets are running. But there's often little or no financial upside for them to improve the returns. And kind of our model that we have even started now, offering to third parties is a little different. We want to be, you know, compensated or measured for the value creation, not just for keeping the asset running, but for actively improving what it earns. So that means, you know, participating in new revenue streams that maybe before we weren't even tracking. It means catching reconciliation efforts. It means curtailing logics and series service participations in a way that continually optimizes and not just maintains the status quo. So what does that mean in numbers? Internally, we are targeting roughly 150 to 200 basis points of IRR uplift across our portfolio through asset management and commercial value enhancement, as we call it. Not through only leverage, not through acquiring cheaper assets, but really through the operational and commercial excellence on the assets that we already own. That's the number with which we justify our existence as a platform. And that's the number we are also offering to third party clients.

Speaker A: Right. Would that also include revenue management? Like, would you be able to split out, uh, the value increase or the value add from dedicated revenue management?

Speaker D: Yes, that's actually part of that, if the clients, uh, are interested in that. The difficult piece is always agreeing on what a fair baseline is. Of course.

Speaker A: Right. What would be a fair baseline, for example?

Speaker D: Or I think you have to make yourself honest and look at like, okay, what is the business case? What are the underlying assumptions? Is this achievable? And what is a fair expectation in the market? And by agreeing on that baseline, we can then work on, of course, outperforming it. But of course, if someone comes to us and says, hey, I need €200 per megawatt hour, uh, from solar in the German market, because that's in the business case, I mean, we also can't do magic.

Speaker A: Uh, I think we could do a separate one or two or three podcasts. Just what is the correct baseline for, uh, a revenue management organization? David, you mentioned this term, uh, digital ipp. So let's shift a bit to AI digitalization. But what is a, uh, digital ipp? What does this mean for us?

Speaker D: It means mainly one thing. Everything that can be automated is automated. And every decision that still requires human judgment is supported by data that's clean up to date and trusted. That sounds very simple, but the implementation is not. Because you need a data architecture where you have this single source of truth that exists and everyone trusts it. You need to have an architecture that allows to ingest raw data, processing, standardizing it and then preparing it for consumption of different reports and tools and so on. And you need process logic encoded in the software rather than in people's heads or in spreadsheets and so on. You really need the discipline to actually build and maintain that over time. And that is of course an investment. The AI layer, so large language models in particular has dramatically accelerated our ability to build these systems. So tasks that would have required dedicated software developers, six months project cycles, et cetera, we are now doing it iteratively, often in weeks and with small internal teams. The key for us, and that is part of our digital strategy, is to empower the people to help themselves. So our people from analysts to the C level are building, uh, prototyping and modifying their own tools and that is fundamentally different in terms of the operating model. Now and then our digitalization team helps to turn self service tools or prototypes into professional software. For example, over the last six months we built a completely new and tailored uh, application for our asset managers holding all the master data, all operational data, as well as uh, all internal and external reports and updates. So they now have this single operating center for asset management.

Speaker A: Maybe let's go a bit deeper into that. So how would that look like just from a um, capability standpoint for an organization, when we talk about the next gen ipps and you mentioned it, you have new teams with dedicated focus on revenue management, uh, on asset management origination. On the short term you have uh, let's say a data team. What does it mean then from let's say capabilities and also third party software which you don't need potentially anymore.

Speaker D: Yeah, in terms of capabilities I uh, think there's a big shift and that's not only happening in energy, it's a shift that is happening across the globe is that you need to be able to leverage these tools because they can if used. Right? Absolutely. Increase your productivity. But of course uh, this is an ongoing change process in every organization. Right. You need to empower the people, train the people, give them access to the tools and also make them understand that you're not looking to replace them. You're looking to kind of get rid of all the nasty and annoying work that they can then automate so that they are freed up to do high quality work. And the intellectual work that maybe is often put aside because of the day to day job that is taking so much time. And what I can recommend is actually that's something that uh, we uh, have kind of tailored to our needs is an AI framework by Gartner which looks at different, different buckets essentially and different maturity states in terms of your AI journey. It basically looks at, okay, what are you doing in Terms of data, uh, governance, uh, training and integration and so on and track certain steps along the way to mature on that AI journey. So that has been quite helpful for us.

Speaker A: And if you were to go practically into one, um, specific field, for example, you mentioned implementation of revenue logic. So you're doing all those deals in the past, you would have maybe bought a third party software for trade capture, revenue risk calculations, revenue tracking. Is this something where you feel now confident to do this fully in house thanks to AI leverage of those tools? How do you think or go about such a specific workflow now?

Speaker D: Increasingly, yes. So, and you asked about numbers earlier, so I'll give you numbers, but I'll also frame it. Honestly, some of this is still work in progress. I'd estimate we're around half the way there. So the numbers we are tracking against are not uh, yet fully realized savings. But on the make versus buy dimension, we've built tooling internally now that were previously paying, that we were previously paying for external vendors or would have needed to pay for. So the combined effect once we are fully there, we estimate to be roughly 1 1/2 million euros of saved one off implementation costs and around €500,000 of annual subscription savings at full scale. So for a company of our size those are really meaningful numbers. But that entails the entire organization, right? So if we look at the, let's say the commercial tech stack, I mean still the heart of it is the data lakehouse. But now using, using also LMMS LNM's ability to translate, you know, language contracts, et cetera into code, we've built, uh, the full revenue logics, uh, internally, um, reflecting all the PPAs, balancing agreements, even now curtailments and redispatch compensations which are kind of automatically calculated based on algorithms that try to capture, okay, was this curtailment based on market action, was this based on grid operator action? And that is then later on automatically reconciled with invoices that we receive, et cetera.

Speaker A: Amazing. I mean if you were to name some software categories or tool categories in the uh, energy revenue space, which ones are affected, like what are you making or insourcing.

Speaker D: So I think given that we are a company that doesn't frequently trade on the market, you know, we thought about having things like an uh, ETRM M, geo management tools, et cetera. So we decided that we can build rather lean things that fulfill the needs that we have, which are of course not the same as a full blown trading company without having to invest heavily into software. Um, it's also, I Think a lot about data collection of what is freely available through portals where you either have access or through public APIs, I mean collecting entity data, et cetera. What is of course hard. And I think that is often, I think the future of software as a service is proprietary data because this is what we still require. And that is not something you can tell an AI to produce or to grab easily. So it's this proprietary knowledge and IP that these companies in the future have to focus on. I think more important than also looking at only cost savings and metrics, I think is also what we can do now that we couldn't do before. And that is the kind of reporting quality and the speed of analysis because we can really tailor things quickly without having to ask for new features that external vendors might have to implement on their roadmap, um, and abilities to really run scenario models at our asset at portfolio level. And those capabilities were previously either unavailable or would have required kind of a team much, much bigger than we would have. Ah, so that's harder to monetize on a single line, but it's a real value driver. And this was only possible, I think, because we had a genuine greenfield advantage. Right. So we are a young company. We had no legacy systems to migrate, we had no existing vendor contracts to unwind, no incumbent IT infrastructure that we were locked to. So we were lucky to be born in a time where we could turn a disadvantage, uh, the lack of all these things into what we think is now actually an advantage because we can move much quicker than I think much larger organizations can at this point in time.

Speaker A: Yeah, it's really empowering, especially for greenfield, smaller ipps starting from scratch. I was just wondering, on your digital team, the IT team, like, what type of people do you employ or would peers like yours need to employ to follow a similar journey?

Speaker D: Honestly, our team is quite mixed. So we have of course, things like, uh, you know, dedicated IT administrator who takes care of our, uh, environment of our hardware, of, you know, IT issues. Then we have, uh, what I would call a business liaison. So someone that translates the business's needs into IT specifications and that is an expert on it, helping people to help themselves and also helping them to then turn prototypes into actual usable pieces of software that scale. And thirdly, and that is super helpful, we have someone that has a background actually in asset management, but is kind of digital enthusiast and coder and that makes life super easy because he knows what the team needs. You don't need to tell him, explain him all the things that drive the complexity of asset management, he knows, and he can just be, build and implement at quite a good speed because the limitation that he had previously, which was the coding capabilities, is now partially taken away and he can really directly implement his subject matter knowledge and build software based on it.

Speaker A: Yeah, we're seeing this everywhere. David, where do you see the downsides or limits to this new model, if any?

Speaker D: I mean, if, if, uh, what sometimes keeps me up at night is a bit the point that while we are really, you know, operating at a rapid speed right now, we also have to make our operations AI resilient. And what do I mean by that? I think that has two aspects. First is a certain provider independence. If your workflows depend only on a specific model from a specific vendor, uh, maybe you're using just a specific LLM harness, you are exposed to risks that you can't fully control anymore. So we had a very specific demonstration of that a few weeks ago when the US government ordered Anthropic to suspend access to its most capable models. And Anthropic had to take them down entirely for a period, including for European business users. I mean it got resolved, but for a window of time, any company that had built critical workflows around these specific models was blocked. So it's not really a theoretical risk anymore. And the question to ask yourself is, if your AI provider goes down tomorrow or gets restricted or raises prices 10 times, does your business still function? Do your processes degrade, let's say gracefully, or do they just collapse? And second is, and this is maybe even the harder of the two disciplines, make sure that your underlying processes and data, data are uh, sound even without the AI layer. So AI for me is a multiplier. If your data is bad, your processes are incoherent. If the underlying logic isn't documented anywhere, um, other than inside a, you know, models context window, then it's, you're not building resilience, you're just automating fragility. So we are big believers in AI at tion, but we run our stack in a way where the model accelerates the decisions that humans still understand and can override the moment where that reverses. So where humans are ratifying decisions that they don't understand anymore because the model said so you have a governance problem and it might not translate into efficiency gains anymore.

Speaker A: Right, right. So another wake up call maybe thankfully early enough for many companies to fill. Uh, direct. David, uh, last question. What's your advice to peers in the IPP and phone space who are at the beginning of this Journey, uh, to

Speaker D: you it would not come as a surprise. I think we talked about it over the years many times. Uh, my single biggest advice is start with the data, not with AI tools, not with vendor selection process, with the question of what data you actually have, how you can get data that you don't have and whether you trust it and where it lives. Is it in one place or is it in 12 different spreadsheets over uh, 12 different people? So once you have that single source of truth, the AI layer is genuinely as powerful as the hype says. But without that foundation you're just generating impressive looking outputs. But on unreliable inputs and maybe to look at this a bit on the commercial side of things, what should IPPs and funds look at? It's challenge your incentive structures, whether ask kind of every counterparty that you depend on, direct marketers, O and M providers, your asset manager, do they have an incentive actually pointing at your returns. If you're paying for effort rather than for outcome, you're not aligned. And that's I think the thread that runs through everything that we talked about today.

Speaker A: Amazing. David, incredible what you are building together with Tyon. Thanks for sharing your journey, your insight. Thank you.

Speaker D: It was my pleasure Luca, thanks.

Speaker C: So let's look at what the team has been reporting and we start in PJM M. PJM is the largest wholesale power market in the United States and possibly in the world. And its problem is not price, its problem is getting anything built. Load in PGM is expected to grow by more than 30 gigawatts between 2024 and 2030 driven mainly by data center expansion. And new supply has not kept up. So the numbers PGM published on 3rd August are worth a uh, look at. Nearly 350 battery storage projects apply to enter the latest interconnection cycle. That's about 60 a 68 gigawatts of capacity. Of those 314 projects qualified, that's just above 60 gigawatts. So roughly 90% of what applied got through. That makes batteries now the leading resource by project count and second only to natural gas by capacity. To put that in perspective, currently installed capacity in PGM is just 600 megawatts, 600 megawatts on the ground, 60 gigawatts in the queue. Now normally a big queue number tells you very little because queues fill up with projects that were never serious. What makes this now different is that PGM has overhauled the process it now runs on a first ready first served basis like in Ercot Texas rather than the old first come first served a process approach and the qualification stage is deliberately rigorous. Also, the economics are pulling in the same direction. Battery contract valuations in PGM have risen alongside record capacity prices and capacity now accounts for about a third of total project valuation. Take a four hour battery seven year term delivery at the Western Hub starting in 2028. This valuation rose by over 20% from the start of the year to now whopping $15.07 per kilowatt month according to Paxo park data. That was for the 13th of August. Now let's see how the clearing of the queue will proceed. But it looks more likely that we will see finally the best boom taking off as well in pjm. Fingers crossed. Second story it's market story Although it starts with regulation, we're coming to southeastern Europe and what the carbon border is doing to prices. We've covered C band before, so briefly. The definitive regime started on the 1st of January this year after a transitional phase that ran for two years or so. And it covers imported electricity to to the European Union from outside its border. What matters commercially is that importer power is generally charged now on a country specific default carbon emission factor unless every condition is met for using the actual embedded emission of the specific generating installation. So what does that mean? Let's look at it six months in. Here is what we saw in prices in the first quarter. The Western Balkans really pulled sharply away from neighboring EU markets. For example, Italy averaged around €44 per megawatt hour above Montenegro. Hungary traded roughly 33 hours above Serbia. And all of those prices, I mean they are a big departure from where those borders sat just a year earlier. Now the second quarter brought some convergence, but not a return to the old price relationships we observed over the years. Now the tempting read is that the frictions went away, but they didn't. A high default carbon cost can make cheaper non European power uneconomic to export into the eu, which weakens the arbitrage and leaves more electricity at home in the non European countries. I mean non European Union countries. But that was just one factor among several. The first quarter also had exceptionally strong hydro, up 33% year on year across the region. But the second quarter that surplus has faded and the western Balkans shifted to net imports which as well supported local prices. Then we had softer demand, stronger renewable output, cheaper gas link generation, all pushing EU prices down. So we could say prices converged because fundamentals moved on both sides of the order of the border, not just because the pre CBAM relationship came back. Now here's the part that matters for contracting cross border PPAs in these smaller less liquid markets are often settled against the neighboring EU price and now with cbam, as coupling weakens, that benchmark drifts away from what the project actually earns basis risk. The Energy community secretariat took 130 megawatt onshore wind project to size the impact and over the first half of this year selling that output domestically rather than at Hungarian day ahead prices would have meant an opportunity cost of about 8.9 million euros. Huge. So the direction of travel is that CBAM could stimulate PPA activity across the Western Balkans and Eastern Europe Europe but the transitions get more complicated. First, a generator outside the European Union doesn't surrender certificates, but if its power is imported, being able to use actual installation emission instead of the country default. This can materially change the economics. But a uh, PPA between the authorized declarant and the non European producer is just one of several comuled cumulative conditions for that and compliance has to be verified by an accredited verifier. Which means the value of a renewable PPA is increasingly dependent not just on the price it fixes, but how cleanly it can link back a specific asset to a specific European buyer. This is dizzying. Life is complicated as always, but it is especially complicated for renewable PPAs in the European Union.

Speaker B: Thank you for listening to the Pexapark podcast. If you're interested in more news, data, uh, insights and analytics on the energy transition, head to our website pexapark.com to find out more. It.