0
This week’s conversation is with Scott Blalock of Australian energy company Wärtsilä. I spoke with Blalock this week amidst my reporting on transmission after getting an email asking whether I understood that data centers don’t really know how much battery storage they need. Upon hearing this, I realized I didn’t even really understand how data centers – still a novel phenomenon to me – were incorporating large-scale battery storage at all. How does that work when AI power demand can be so dynamic?
Blalock helped me realize that in some ways, it’s more of the same, and in others, it’s a whole new ballgame.
The following chat was lightly edited for clarity.
So help me understand how the battery storage side of your business is changing due to the rise in data center development.
We’re really in the early stages for energy storage. The boom is really in generation – batteries aren’t generators. They store, they shift, they smooth power, but they don’t generate the power from fuel. In this boom right now, everyone is trying to find either grid connections or on-site power generation. Those are the longest lead time items ± they take a while — so we’re still in the early stages of those types of projects coming back and saying, we need to start procuring batteries. We need to start looking at the controls and how everything’s going to work together. That’s still a little bit in the future.
Are you seeing people deploy batteries responsibly, in an integrated way, or is it people unsure what they need?
There’s definitely uncertainty as to what they need. The requirements are still hard to nail down. A lot of the requirements come from the load curve of the AI workloads they’re doing, and that’s still a bit of a moving target. It’s the importance of knowing the whole system and planning that out in the modeling space.
The biggest space of all this is the load profile. Without a load profile, there’s uncertainty about what you’re going to need –
When you say load profile, what do you mean?
The AI workload. The GPUs. The volatility. In a synchronized training load, all of the GPUs are generally doing the same thing at the same time. They all reach a pause state at the same time, and you’re close to full power on the data center, and then they say, okay now we go idle. It has a little bit of a wait and then starts back up again.
It’s that square wave, very sharp changes in power – that’s the new challenge of an AI data center. That’s one of the new uses of BESS that’s being added compared to the traditional data center doing data storage. They’re more stable which use less power and are more stable.
The volatility is where some of the friction comes in, and that has to be handled by some technology.
So what you’re telling me is that data center developers do not know how much they need in terms of battery storage? Simply put, they don’t know how much power they need?
Traditionally, utility-scale batteries – the projects we’ve been doing – come from a PPA, an interconnect agreement. There’s something in place where they know exactly how many batteries they can install. They know how many megawatts they’re allowed to install. Then they come to us and they say, I need a 4-megawatt battery for two hours. Tell me how many batteries you’re going to give me.
In a data center, they don’t know that first number. They don’t know how many megawatts they need. So that’s the first question: well, how big of a battery do you need?
If you have a 1-gigawatt data center that means the load change is 60% of that – 600 megawatts is the step up-and-down. The starting point is 600 megawatts for two hours. That’s the starting point that’ll cover being able to take care of that volatility. The duration is a part of it, too. From there you get into more detailed studies.
When it comes to transmission, how much of a factor is it in how much storage a data center needs?
The first thing is whether it’s connected at all. The battery is a shock absorber for the whole system. If you are grid-connected, the BESS is still a stability asset – it’s still improving the power quality and stability at an interconnect. If you’re doing on-site generation, it becomes vital because you have only one system being controlled.
As far as when you talk about permitting and transmission, the details of that don’t really play that much into the BESS, but it’s tangentially related. The BESS is an important part of how you handle that situation. Whether you get to interconnect or not, it’s an extremely important asset in that mix.
With respect to the overall social license conversation, how does battery storage fit into the conversations around energy bills and strain on the grid?
Bias aside, I think it’s the most important piece.
If you look at the macro scale, it’s like transitioning to renewables where they’re intermittent; batteries turn intermittent generation from renewables into firm, dispatchable power. It’s still not going to be available all the time – you’re not going to turn a solar plant into a 24-hour baseload plant – but a battery allows you to shift the energy. It greatly alleviates the problem.
The other aspect is it’s a stability asset. The short version of that is you have big thermal plants – rotating metal masses that have momentum to them that stabilize everything on the grid. As you take those offline, the coal plants and the gas plants, the grid itself loses that inertia so it is more susceptible to spikes and failures because of small events. Batteries are able to synthesize that inertia.
