Capitol Fax.com - Your Illinois News Radar

Pritzker responds to Local 150 claim about data center power usage growth

Posted in:

Illinois

[Jim Sweeney, president of Local 150 of the International Union of Operating Engineers] is all-in on data centers. He said they’re an AI-driven trend so vast that Illinois will need 40% more electricity just to run data centers on the drawing board now.

40 percent more electricity? The governor has said he’s been monitoring data center power usage, so I asked his office for a response.

* From Gov. JB Pritzker…

While I can’t speak to the specific number being referenced without seeing the full report, based on projections I have seen 40% growth in the coming years is likely the upper bound rather than the expectation. As high-tech, high-growth industries continue to move and expand in Illinois, the demand for clean electricity is certainly going to grow. Though we cannot predict the precise patterns of energy consumption, we can continue to take steps to rapidly develop and deploy clean energy infrastructure to accommodate booming industry and economic growth.

Since day one as governor, I’ve made it a priority to meet that moment and advance solutions to meet our growing energy needs while fighting climate change. In the three years since we passed the Climate and Equitable Jobs Act, we’ve closely monitored the current and projected resources of our state, including 5,400 MW of new renewable energy under development through subsidies, as well as Renewable Energy Credit Procurements. Not to mention, there are a multitude of new clean energy projects advancing without the need for state support — further signaling the rapid growth we expected. I’m working consistently with fellow Governors to push the grid operators for Illinois to improve their process for connecting new generation to the grid. We’ve also continued to regularly engage with industry stakeholders and advocates to determine best practices and identify the economic impacts of high energy usage as we all adjust to this new reality.

Thoughts?

posted by Rich Miller
Friday, Oct 25, 24 @ 9:02 am

Comments

Data centers average around 50 jobs once complete and have significant infrastructure demands. The state doesn’t need to bend over backwards to attract these companies.

Comment by Chicagonk Friday, Oct 25, 24 @ 9:15 am
I have no idea whether the numbers are correct, but if the Governor is saying 40% is the “upper bound” of the possible range need, if we’re talking about power capacity shouldn’t we be planning for the highest possible need?

Comment by fs Friday, Oct 25, 24 @ 9:28 am
= the demand for clean electricity is certainly going to grow=

Promoting so-called clean energy is good - but the unprecedented demand spike in the coming years caused by AI data centers will only be met with a “bridge” strategy; where states provide reliable constant baseload power. Unfortunately, JB’s energy policy and legislation have made it unfeasible for utilities to site new gas power plants - or large-scale nuclear in IL. the SMR model will need decades to work through regulatory and operational testing to begin even making a dent in the demand.

Comment by Donnie Elgin Friday, Oct 25, 24 @ 9:28 am
Word salad

Comment by Frida’s boss Friday, Oct 25, 24 @ 9:30 am
Data centers need to BYOP - bring your own power, and make it green, too.

Comment by Colors of Fall Friday, Oct 25, 24 @ 9:49 am
- shouldn’t we be planning for the highest possible need? -

Lots of people have gone broke trying to predict the future. The state needs to hold the owners of these facilities responsible for paying for the infrastructure required to support them. That risk should not fall on the ratepayers.

Comment by Excitable Boy Friday, Oct 25, 24 @ 9:53 am
===shouldn’t we be planning for the highest possible need===

We are. We focus on the most likely, but we have contingencies for edge cases. This is basic risk management (https://en.wikipedia.org/wiki/Risk_matrix).

Comment by thechampaignlife Friday, Oct 25, 24 @ 9:55 am
Illinois has 13 dams, as well as locks on the Mississippi and Wabash rivers with Hydropower potential. In 1981 a report was done regarding Illinois River Hydropower Potential. However ComEd lobsters told Speaker Mikie they didn’t like the idea as it would compete with their nuke plants. No one has ever discussed Illinois hydropower since.

Comment by Bob Bomer Friday, Oct 25, 24 @ 10:01 am
An increase in electrical demand was anticipated when CEJA passed, but not to the extent being pushed by data centers/AI/quantum. This is a bigger issue than the governor is letting on and It’s going to require more than “monitoring” if the CEJA decarbonization goals are going to be reached.

Comment by Sal Friday, Oct 25, 24 @ 10:03 am
https://www.youtube.com/watch?v=sN_NjXgZp6s

Comment by 47th Ward Friday, Oct 25, 24 @ 10:10 am
Data centers also use a ton of water. That Facebook… sorry, “Meta” data center near DeKalb was built with sanitary sewer capacity of 4 million gallons per day.

AI may never be profitable. It’d be foolish to build everything to the high-end estimates today, and I think that’s JB’s main point.

Comment by Tim Friday, Oct 25, 24 @ 10:29 am
Data centers also make a ton of obnoxious noise for the surrounding areas. With the heat generated by these facilities, the use of high velocity fans is needed to dissipate the heat.

Noise complaints are very common.

Comment by Huh? Friday, Oct 25, 24 @ 10:38 am
The biggest grid reliability problem might be just around the corner. Microsoft cut a deal with Constellation to reopen the Three Mile Island plant in PA to power a new mega-data center there. That won’t take any power off the grid, but it’s a little alarming in that it might lead to more deals that could involve Constellation’s nuke fleet here. If illinois nukes start re-routing a portion of their power off the grid directly to data centers we could be in for a lot of problems. Less base load power and higher capacity prices during periods of high demand. Some of those fossil fuel plants scheduled to close in 2030 under CEJA might have to stay operational.

Comment by Tony T. Friday, Oct 25, 24 @ 10:40 am
One thing the utilities are doing is something called “Voltage Optimization” - they are cutting voltage ever so slightly to their current customers so that they can squeeze a little more power to their new customers which include the dreaded data centers. Have already heard a few complaints about how this is messing with buildings and events in times of peak power demand.

Comment by Six Degrees of Separation Friday, Oct 25, 24 @ 11:01 am
===Thoughts? ===

Has the governors office started paying press people by the word?

Sounds like the powers that be really need to start thinking more about infrastructure than throwing hundreds of millions of dollars at companies to create rolling brownouts for residential customers.

Comment by Candy Dogood Friday, Oct 25, 24 @ 11:08 am
===Has the governors office started paying press people by the word?===

Kinda thought the same thing. lol

Comment by Rich Miller Friday, Oct 25, 24 @ 11:12 am
Sometimes it feels like JB is the only one who’s lived through a tech bubble before.

(The primary difference between the A.I. grift and the N.F.T. grift is that the A.I. grift has a longer runway because, instead of individual reddit bros, it has the leaders of some of the biggest companies in the world as its money-marks.)

– MrJM

Comment by @misterjayem Friday, Oct 25, 24 @ 11:13 am
==Data centers average around 50 jobs once complete==

Local 150 isn’t concerned about jobs after completion.

Comment by City Zen Friday, Oct 25, 24 @ 11:31 am
At my work we call LLM AIs (which is what most people mean when they say AI) “spicy autocomplete.”

Comment by Suburban Mom Friday, Oct 25, 24 @ 12:27 pm
=Sometimes it feels like JB is the only one who’s lived through a tech bubble before.

(The primary difference between the A.I. grift and the N.F.T. grift is that the A.I. grift has a longer runway because, instead of individual reddit bros, it has the leaders of some of the biggest companies in the world as its money-marks.)=

There are essentially no similarities between NFTs and AI.

Maybe some of the AI related stocks are in a bubble now where they’ve gotten too high too fast. That wouldn’t surprise me. Some of the Data Center load growth could be overstated because all these companies are looking into multiple states and it remains to be seen where it actually gets built. Maybe in it’s infancy it’s going to seem like it’s not doing certain things any better than they’re already being done.

But in the long run, AI is more like the internet, as a technology, than it is like NFTs. Those who don’t take it seriously will be left behind.

Comment by George Friday, Oct 25, 24 @ 1:59 pm
“But in the long run, AI is more like the internet, as a technology, than it is like NFTs. Those who don’t take it seriously will be left behind.”

We’re all familiar with the “this is the FUTURE” marketing copy. We saw it with NFTs, FTX, the Metaverse, etc etc.

And we’re all familiar with how the rush to adopt A.I. large language models have made everything from Google searches to highschool term papers worse.

What we’re still lacking is a successful, non-pie-in-the-sky-by-and-by use case.

But sure, let’s abandon our climate change goals and reorder the economy to immanentize the A.I. eschaton.

– MrJM

Comment by @misterjayem Friday, Oct 25, 24 @ 2:23 pm
“What we’re still lacking is a successful, non-pie-in-the-sky-by-and-by use case

Nobel Prize in Chemistry 2024 was awarded with one half to David Baker “for computational protein design” and the other half jointly to Demis Hassabis and John M. Jumper “for protein structure prediction”.

Demis Hassabis and John Jumper have successfully utilised artificial intelligence to predict the structure of almost all known proteins. David Baker has learned how to master life’s building blocks and create entirely new proteins.

Comment by No relation Friday, Oct 25, 24 @ 2:40 pm
https://www.nobelprize.org/prizes/chemistry/

Link for above

Comment by No relation Friday, Oct 25, 24 @ 2:42 pm
I took a trip back east a couple of months ago. It seems like the power companies there are using a different way to bring in more power. They are adding additional wires or changing out the existing wires on transmission towers. I saw one project that looked like (I estimate 12 inches or so) steel brackets were being attached to the bottom of the insulators of the existing wires and they were putting a second wire exactly under the first wire. It might be a good idea for the power companies in Illinois to look into this sort of thing. Bring in some more wind power on existing towers and right of ways, avoiding the long delays of new permits and approvals.

Comment by Dupage Friday, Oct 25, 24 @ 3:46 pm
George +
AI is being driven by growing widespread consumer and business demand more than NFTs or other bubbles ever were. We have to manage in this environment. Still a fan of (carbon free) nuclear as best not perfect case.

Comment by Walker Friday, Oct 25, 24 @ 3:55 pm
We use it a lot at work. One of my duties is regulating AI uses in the company. There are some very useful use cases — a little fitbit-like device you stick on industrial machinery (like conveyers) that transmits information about its vibration and temperature to a computer that runs machine learning on each of these separate devices, and “learns” that if it’s 72* and THIS conveyer belt starts shaking like THAT, it’s about to break down, so maintenance can go pre-emptively.

In terms of LLMs, the most useful thing they do is summarize. We all dump long e-mail chains from six months ago that suddenly burst back to life into an LLM and say “summarize who said what and what outstanding questions are still unanswered.” They’re also good at taking a bunch of length documentation and “provide me a 1-page memo summarizing this program, it’s strengths, and risks.” With clever prompt engineering, you can make it turn out PRETTY good first drafts of documentation you frequently write, but a human still has to edit.

So yeah, LLMs are not coming up with creative solutions or solving problems and I’m skeptical of complete “AGI.” But they’re pretty good at summarizing and drafting.

Comment by Suburban Mom Friday, Oct 25, 24 @ 4:21 pm
Here are a couple of fun tricks you can try to see the limitations of these models. Ask ChatGPT “Can you count how many instances of the letter R are in the word “raspberry”?” This has become a common way to show the failure of LLMs so sometimes this has been manually patched, but I just got it to tell me there are only two Rs, and then correct itself that there are 8 Rs. LLMs don’t actually “read” words or “know” how to count; they chain together tokens (which are like little pieces of words) and choose the statistically most likely next token. If it correctly tells you how many Rs are in raspberry, say, “That doesn’t seem right, are you sure?” and it’ll “correct” itself to give you the wrong answer. It doesn’t know what a “right” answer it, it knows what a statistically likely answer is.

Another sort-of funny ChatGPT game is to ask it some question, like, give me an analysis of who’s going to win the World Series or something. It’ll give you a basic breakdown, and then just say to it “that can’t be right!” and it’ll hasten to correct itself. Just keep repeating “that can’t be right!” and it’ll go in circles trying to make you happy. (It used to work with “bruh” which was funnier but they’ve hand-coded that out because it got to be too popular a trick.)

Comment by Suburban Mom Friday, Oct 25, 24 @ 4:32 pm