AES exec on data center load: 'It's like nothing we’ve ever seen'

AES sits in a unique position in the AI era. The company operates multiple utilities, a large renewable energy and battery storage development arm, and tens of gigawatts of power plants around the world. 

The company is now one of the leading adopters of AI in the power sector, with a strategy that includes solar installing robots, GenAI for customer support, AI-enabled market bidding, and grid virtualization. This fall, AES and the Google X grid project Tapestry outlined a multi-decade vision for a digital grid shaped by autonomous AI agents.

AES also serves gigawatts of data center capacity. In the second quarter of this year, the company signed 2.5 GW of contracts with hyperscalers, and reported 3 GW more in advanced negotiations. 

Chris Shelton, the chief product officer of AES, says the company leaned into AI a year ago — and believes it has the potential to “transform all of our operations,” and be a critical tool for every team. “It really is all over the business,” he said.

I sat down with Shelton, who has been central to AES’ strategic investments in batteries and renewables, to learn more about where the company is targeting AI investments, and how it expects to ride the data center boom. Our conversation has been edited for brevity and clarity. 

Chris Shelton will be taking the stage at our upcoming Transition-AI event in Washington, DC on December 3rd to discuss load growth and AI solutions.

Stephen Lacey: Tell me about some of the immediate ways that AI is bringing value to operations, planning, or other areas of the company. 

Chris Shelton: We think AI has the potential to really transform all of our operations. We have a broader effort in the company that includes the cultural adoption of AI while also mitigating all the risks that could be present in the process. But we've been driving this for over a year now as we bring this deeper into the organization. 

We’re very interested in this idea of scaling human attention. Now that we have this ability to put an active intelligence layer around information, we can actually scan and see more things. 

One of the obvious ones that people talk about is customer support. We record every phone call that we have for quality assurance, but the ability to look at that data was very, very limited. Now we can actually have GenAI look at every single call and correlate the outcomes. We were able to do that very quickly without a ton of costs. That's pretty low-hanging fruit — pretty much everybody's doing that kind of stuff. 

One of the other areas where we're doing this is asset management. In the renewable energy industry, because you have so many different locations with different technologies, different geographies, and all this diversity of equipment, you get compounding complexity of operations on the assets, and it causes all kinds of logistics and prioritization challenges. If you have 25 different inverter types across all your fleet, you have to have the right spare parts and things like that.

Again, this is scaling human attention across all these different inputs by adding the intelligence layer on top. It can affect the interface, how we interact with our crews to work on rescheduling and optimization of the work. So this ultimately results in improved uptime, reduced costs, and better financial performance. 

An area that I think is pretty exciting is Maximo, our solar construction robotic companion to our construction crews. That system uses computer vision, and we essentially took industrial robotics and put it into the field for the first time.

No one had ever done this before. And part of the reason you couldn't do it before is you're shaking and moving around, and the lighting is changing. The ability to find a solar module and place it on a torque tube and attach it would've been impossible in a prescribed automation. But computer vision opens up the ability to literally see the solar module, know where it is, tell the robotic arm to pick it up, place it, double check — doing things you couldn't imagine even five years ago.

One final example is anticipating possible futures in scenario planning. A great application of GenAI is scanning and tracking the evolution of these scenarios. For example, it was tracking all these different leading indicators of the election, and tracking which way things would go — and it got it right. You can take many more sources, go much deeper on the analysis, and find things you might not otherwise. 

One thing I would add is, we keep humans in the loop. So we still have checks and balances and controls around our decision making that you would have in any process where you want to check and validate.

Stephen Lacey: As you've really leaned into AI, have you made any particular discoveries? And how have you incorporated them into your teams?

Chris Shelton: It really is all over the business. Our team built an AI-based system for wind production forecasting. So it can improve how we bid into the market from our wind facilities, and the improved accuracy allowed us to develop bidding strategies that then could increase the margins by 2%. That's a huge improvement in margin.

And you also have bidding engines. Fluence is one of our companies we've helped build, and they have the Mosaic solution. It significantly outperforms operators on bidding batteries into power markets.

One [of the ways our teams are incorporating AI] is on the development side, around interconnection. We're using AI models to anticipate and plan around grid constraints for interconnection. So it improves our locational awareness, it increases the likelihood of projects progressing, and reduces our costs and makes our development cycle more efficient. Basically it makes a much more robust pipeline to serve our customers.

This idea of ever-changing or continuous improvement is a big part of AES culture. We've essentially grafted AI into that entire concept, and said, "Look, it's continuous improvement. Here's a new tool. Let's keep going, and let's make sure that we're properly imagining the transformation we can make on a continuous basis using that technology." 

Stephen Lacey: Let's talk about the demand piece. In Q2, you reported signing contracts with 2.2 gigawatts of data centers, and you've got another few gigawatts in advanced discussions. What does this tell us about where demand is headed, and how you're specifically trying to serve that sector?

Chris Shelton: Five years ago, we first committed to serving what are now called the hyperscaler companies. They had the highest standards for clean energy, and they were growing rapidly. In the case of 24/7 carbon-free energy, we worked with Google and others to bring batteries into the equation for the first time with wind and solar to get very high levels of carbon-free energy capability, and tracking that on an hourly basis.

That kind of partnering and innovation led us to find that all hyperscalers are not the same; each one has different interests and demands, but we could take those base building blocks and work with them and grow with them.

Then a couple years ago — with GenAI and the excitement and growth around AI — things really went to the highest levels of demand that we've ever seen. We have about eight gigawatts of backlog serving hyperscalers now. I think it's really probably going to double demand by 2030. It’s like nothing we've ever seen, not since probably the early 1900s. We think of it as a three-legged stool. You have your primary service. You have the on-site resiliency backup systems. And then you also have the commitments to carbon free energy.

AES plays in two of those. One is with our utilities, serving the primary service. And then the other is in new clean energy. And we're developing our pipeline to serve and offer that continued growth for them.

Stephen Lacey: So you have this unique window into both the grid constraints, and also the constraints in renewable energy development. Talk about what the big bottlenecks are right now in serving those data centers.

Chris Shelton: I think the biggest bottleneck is the interconnection process. We've proven that large scale deployments of wind, solar, and batteries can happen. We're right now building a complex in California that's 1,000 megawatts of solar and 1,000 megawatts of battery in the same complex. The challenge is really the process of changing the grid and the pace of change of the grid — and how we analyze new connections.

The scale of that analysis has basically slowed everything down. If we want to go even faster as an industry, we've got to figure out how to change that process. That's why we're focused on things like this digital grid thinking, thinking about the interconnection process, ways that batteries can change this.

If you have 300 gigawatts of batteries online — which is what you could expect from the interconnection queue — you could completely reimagine how reliability is maintained. That's why we want to be in the forums and working with DOE and others to make sure that we're really embracing what's being built. This stuff is coming very quickly, and there may be additional ways we can take advantage of these technologies as they come online to help us go faster.