This article is sponsored by Ardian
Headquartered in Paris, Ardian is a private investment house with $169 billion under management, and AI is central to its infrastructure investment process. The company sees its capacity to invest in teams of digital and data science experts as a key element of the benefits the asset manager brings to portfolio companies.
Marion Calcine, chief investment officer and senior managing director – infrastructure, and Pauline Thomson, head of data science and managing director – infrastructure, explain why AI is a trump card for managers and how its development will affect the future of infrastructure.
Can AI accelerate the energy transition in infrastructure, and which are the most important tools?
Pauline Thomson: We see a direct correlation between operational excellence and impact on carbon footprint. We use a lot of data-driven tools to help the energy transition.
For example, we have built a data-driven model that allows us to calculate the carbon footprint of our transport assets. Our Ardian AirCarbon platform is deployed in several airports, including Milan.
An airport is able to share with us all their traffic data and some of their operations data linked to ground service equipment, and we share with them the outputs of these models through automatic dashboards that they have access to, and different simulations. It connects directly to the traffic databases and computes all the carbon emissions that are linked to landing and take-off, taxi time, ground service equipment and access to the airport, so it allows airports to have a granular and dynamic view of their carbon footprint.
We also build forecasts and determine what an airport can do to decrease that footprint. There are several levers such as decreasing taxi time, including sustainable aviation fuel in the energy mix, or incentivising airlines to use efficient aircraft.
How else can AI be used to increase sustainability?
PT: We have also built a platform called Opta for managing renewable assets. That tool allows us to both optimise our operations and our revenue profile.
We collect 800 signals every 10 minutes from all our wind and solar assets and from this data we run automatic analysis on operations to make sure that everything is running properly. The tool generates alerts if we see that there is an underperformance somewhere in our fleet.
Then on the market side, we are able to simulate contracting strategies, both back testing and forecasting, depending on different PPA structures, or different hybridisation strategies.
Marion Calcine: From the asset owner perspective, data analysis tools enable us to optimise performance. We are able to identify portfolio-wide patterns which management teams locally on the ground would not necessarily be able to identify.
How else can an asset manager use data models to gain an advantage?
MC: We bring to management teams data science expertise that they would otherwise not have internalised. We bring them anonymised operational data on a portfolio much bigger than the portfolio they manage themselves, helping them improve performance.
Opta also helps them assessing the incremental return and the reduction of risk that diversification can bring, for instance when colocating a battery storage asset to an existing solar plant.
Finally, we create a community of CEOs where each of them can exchange ideas and benefit from cross-fertilisation. That makes CEOs are keen on working with us, because they see that we bring more than just financing.
Tools such as Opta also generate competitive advantage for us as asset managers, because they gather all sorts of data which we can then use in acquisition due diligence. We can sometimes factor in improvements in performance that we will be able to reach when calculating pricing.
More broadly in portfolio management terms, data models such as these are part of the comprehensive strategy we have for renewables, which is long-term investment and optimisation of the renewables fleet through digital tools. You are able to create more value because you create more diversification, in terms of technologies and price zones, and, as a consequence, more stable volumes.
Finally, we can gain advantage by using data analysis and data science for portfolio construction. We are building models which enable to run ‘what if’ analysis. We project macroeconomic variables such as interest rates and inflation and we look at the impact on return, cash yield and volatility of adding a given asset to a portfolio.
What are the primary challenges with AI and how concerned are you by the power and resources required by the technology?
PT: When we own data centres the first thing that we look at when we make an acquisition is connection to the grid and ability to be provided with power. And for the data centres we own, we discuss on a continuous basis with the grid operators to make sure that they will deliver on the additional power that we need.
Of course there are other challenges. One of the key ones for generative AI is hallucination. You need to always have a human to make sure there is no hallucination in the answers you are getting. You need strong internal governance when you deploy this kind of tool.
More broadly speaking, there are challenges with AI coming from increased cyber-threats. There are more of them, and they can be more powerful than before.
But, yes, with all of that said, the main challenge is the broader societal one around the power needs of AI. Power is really becoming the bottleneck for AI and even for data centre deployment.
If we look at data centres, AI compute is supposed to grow at 15 percent CAGR globally from now to 2030. That means that data centres would account for more than 7 percent of worldwide energy consumption, compared to just 1.5 percent today. Society needs to invest very heavily, both in power generation and transmission.
This makes it no wonder that so many hyperscalers are announcing massive plans in the Nordic countries where you have ample power supply and renewable energy, and of course naturally lower temperatures which makes the data centres much more efficient. And this is why we have recently closed the acquisition of Verne, a data centre platform focusing on the Nordic countries and providing sustainable power for high density AI compute.
What of the future, could AI go further than data gathering, perhaps taking on analytical responsibility?
PT: The way we approach generative AI is that it is here to assist, but never to replace human expertise. We have built an internal application called Gaia which leverages on the large language models that are developed by OpenAI and Mistral. Analysts can upload a folder of documents and Gaia will search for them and extract data, textual data, information. It can compare documents, it can summarise, but it cannot replace industry knowledge and analysis.
AI tools do not contextualise the information, so it is purely a statistical model. In the end, you cannot rely on these tools to take decisions that require experience and critical thinking.
We have looked at a lot of predictive maintenance use cases in the past, and they have never fully fulfilled their promises. If we could build more accurate predictive maintenance models that would have a meaningful impact on operations, that would be a considerable leap forward.
How do data models affect value creation? Have you measured it?
MC: One very efficient measure we have implemented is the return on investment provided by Project Opta. We calculate Opta adoption has led to increasing the average time-based availability of our wind farms portfolio in Europe by 1.6 percent, meaning that our wind farms are ready to produce six more days per year on average.
We estimate we actually gained average yearly revenues of €10 million on our wind farm portfolio. And as these revenues come with no additional cost (wind and solar are free resources), this is something that goes directly to the cashflows of the companies.
But the value created is also linked to the behavioural change that our management teams go through. When you provide them with a unique digital tool which tracks performance, they are willing to track, monitor and improve.