Analysis
Sponsored
DATA + CLOUD
AI

What does it take to make hourly matching work for a single data center?

Inside Gridmatic’s effort to “scale down” time matched energy for smaller operators.

|
Published
March 22, 2024
Listen to the episode on:

Photo credit: Ou Dongqu / Xinhua via Getty Images

Photo credit: Ou Dongqu / Xinhua via Getty Images

Data center owner and operator EdgeConnex is using a portfolio of options — renewable energy credits, short-term power purchase agreements, energy supply contracts, for instance — as it works to meet its goal of zero carbon energy by 2030. 

But over the course of the last year, the company has been experimenting with another, less widely relied-upon, option: hourly matching powered by artificial intelligence.

  • The top line: In partnership with the power marketer Gridmatic, EdgeConnex has been running an hourly matching pilot at its 93,000 square foot data center in Houston, Texas. Relying largely on local wind power, the pilot reached 80% carbon-free energy usage in real-time, hourly increments. While the project’s scope is small so far, it bodes well for the potential of applying advanced technologies to hourly matching, especially as the energy industry scrambles to prepare for the impending data center load spike.
  • The current take: Despite the intersection of corporate decarbonization goals and load growth concerns — which are acute enough in some contexts to keep coal plants online past their retirement dates — hourly matching for individual data centers still isn’t widely available, said David Miller, Gridmatic’s vice president of business development.While major players like Microsoft and Google have been working on similar solutions, they would be challenging for smaller companies to imitate: “If you look at a single facility, it’s smaller, and this is a problem that is a little bit hard to scale down,” he said. 

Chief power officer Raj Chudgar said that EdgeConnex’s colocation model — allowing multiple customers to rent space in a single data center — the energy needs and emissions of its centers are different from those of Big Tech.

And long-term PPAs aren’t a viable option for all of its data center locations, in part due to the transiency of the company's customers, he added. That’s where Gridmatic comes in: using its prediction software to provide optimized energy portfolios on behalf of customers that don’t want to get locked into 20 year PPAs.

In operation since 2014, the company’s Houston data center has a very flat load, Chudgar said, making it the ideal guinea pig for a long-duration pilot. That particular center has around a 90% load factor or capacity usage, making it relatively easy to know hourly load.

To start, Gridmatic set the hourly matching target for that data center at 60% — meaning that 60% of the time, the data center’s energy consumption would be matched to renewable energy every hour. That’s slightly higher than the 40% overall penetration of renewables in ERCOT, Miller explained.

That was what Miller describes as a “key decision”: rather than an ambitious, 100% renewables target for 20 years from now, the team prioritized a near-term, more easily realized target. 

Gridmatic was responsible for building a portfolio of renewables to fit the needs of the Houston data center and deliver time-matched energy, he added. The goal? To help the data center rely on a higher share of carbon-free power than they’d otherwise have from the existing grid mix in the region.

To meet the initial 60% target, Gridmatic considered a variety of options, but ultimately concluded the easiest place to start would be with off-site power contracts with wind farms. 

“We expect that over time it may grow to include solar and storage and potentially other resources, but getting those initial low-cost resources was the first step,” Miller said. “We looked at a target that we could achieve relative to the existing grid in ERCOT, where the mix is something like 40% carbon-free, including renewables and also nuclear.” 

To date, the pair have reached an average of 80% time matching, with some seasonal variation.

But can it scale?

For EdgeConnex, the top priority for the pilot was to determine whether hourly matching could be easy enough to expand across the company. 

“If we can’t scale it, we shouldn’t do it,” Chudgar said. “Part of the reason we set a 60% target was that we didn’t know what we were getting into. “We wanted to get to the isotopic maximum that allowed us to get maximum renewables on a time matching basis without doing gymnastics, and then afterward look at additional time matching certificates to get to 100%.”

Those “gymnastics” would include options like adding behind-the-meter supply or load shifting programs, which have been widely experimented with across the data center industry, but which have yet to become the norm.

Gridmatic — which was founded by a former Google data scientist — uses proprietary AI to model and forecast energy supply and demand, as well as pricing. But, as with nearly all AI endeavors related to power markets, there were a few problems when it came to the data.

For load data, for instance, the team primarily utilized the data center’s smart meters, but those have a one day delay, Miller said, causing data availability challenges

“You could use more sophisticated instrumentation, but we wanted to try to use smart meter data as a more scalable approach to other locations,” Miller added.

There were also delays with generation data.

According to Miller, most third-party generators can’t share what they’re producing in real time because of power market regulations: “Because we don’t have live data for any of these things, if we want to utilize any kind of flexibility we need to do forecasting for what the facility is actually going to be consuming and what the renewable generation is actually going to be producing.” 

And that’s where Gridmatic’s specialty lies, he added: advanced forecasting for specific customer verticals.

Listen to the episode on:
No items found.
No items found.
No items found.
No items found.

Beyond Houston

Chudgar said EdgeConnex will be keeping a close eye on how the percentage of hourly matching shifts over the summer of 2024, which, given the extreme temperatures in Texas, is a “much more interesting season” than any other time of year. 

(That’s in part because the Houston data center is air-cooled, which tends to be more energy intensive than water cooling, particularly in warmer months.)

Moving forward, Chudgar said EdgeConnex is deciding which data center they’ll bring Gridmatic’s hourly matching program to next.

Meanwhile, Gridmatic is eyeing its technology’s longer-term potential. In the coming years, there are several industries that are likely to require hourly matching, Miller said, so now is the time to iron out snags such as data availability.

For example, under the current proposal for 45V tax credit eligibility, hydrogen companies looking to take advantage will have to hourly match their green production with renewables in the coming years.

“The work that EdgeConnex is doing at this stage is voluntary,” Miller said. “With hydrogen, and the tax credit, this concept is going from something that’s purely voluntary to something that’s required in order to be eligible for the tax credit.”

Outside of hydrogen, Miller said that impending updates to the World Resources Institute’s Greenhouse Gas Protocols, a framework for measuring emissions, could create another category of companies that would require hourly matching services.

“Those are in the process of being rewritten right now,” he said. “And there’s lots of discussion on whether all companies will soon be looking at time matched energy attribute credits as opposed to annual matching in terms of their carbon reporting.”

And all of those things are happening in parallel: “These different industries are looking at each other to standardize and create established approaches, because they’re all trying to go in the same direction,” he said.

Want to know more about how the spike in data center energy use is impacting the grid? Sign up for Latitude Media’s next Transition-AI event, upcoming on May 8. Latitude Media Executive Editor Stephen Lacey will be joined by three experts who offer a range of views on how to address the energy needs of hyperscale computing, driven by artificial intelligence: Brian Janous (co-founder of Cloverleaf Infrastructure and former VP of energy at Microsoft), John Belizaire (CEO of data center developer Soluna), and Michelle Solomon (senior policy analyst at Energy Innovation).

No items found.