Hilliard's decision to use Electric Aggregation saved every ratepayer about $300 last year. Guess what Ohio's decision to become the Data Center capital of America is about to do?
My last professional role before retiring in 2000 was as the VP/General Manager of the Global Web Hosting business for UUNET, then a division of MCI Worldcom. I cut my teeth in the Computer Technology division of CompuServe, which was a group which built our three data centers (Upper Arlington, Dublin and Hilliard) and operating the large number of mainframe computers on which we ran the products used by our customers. The data center we built at 5000 Britton Pkwy (now the Verizon building) was one of the largest in the region. Today, it's not much larger than the power management rooms in the hyperscale data center buildings Amazon Web Services has built next door, each building 10x the size of our data center - and there are 5 of them on this site. And 5 more in Dublin. And 5 more in New Albany. Another batch on Cosgray Rd are nearly finished, and yet another on Scioto Darby is well along.
We picked the site on Britton Rd for exactly the same reasons AWS did: Access to plentiful electric power (we connected to two substations for redundancy), easy network connectivity (again, to two separate Ohio Bell Central Offices, plus our own private fiberoptic loop between the three data centers), cheap land, and a friendly local government.
I loved it when the first datacenters started appearing in Hilliard. AWS worked a sweet deal with the school district to pay, in lieu of property taxes, somewhere around $1.25-$1.50 per square foot annually to the school district (Norwich Township got left out of that deal of course). This will become a massive revenue source for Hilliard Schools, and should be immune to the threat of the elimination of property taxes (unless AWS anticipated that possibility and has an escape clause in their revenue agreement). Best of all, those data centers would put little demand on local resources: not schools or the police or the fire department. No traffic on the roads to speak of when construction is complete. No pollution other than the rare occasions when they have to run their diesel generators (and they have a LOT of them).
But it's getting out of hand - here an nationwide. The drive to built more and more inconceivably large neural networks to support nano-targeted marketing and a myriad of commercial and industrial applications has led to a whole real estate sector focused on finding greenfields anywhere in the country (planet) capable of supporting data center campuses which consume as much electricity and water as a decent sized city. Check out an outfit called "Data Center Hawk" as one example.
This frenzy sits upon on a well-known set of algorithms for training neural networks that were first developed decades ago. But these algorithms require an intensity of calculations that were not economically or practically feasible given the computing systems of the time. But then someone figured out that the graphics cards - GPUs - being produced by folks like nVidia to enable more and more realistic experiences for the gamer universe were exactly the architecture needed to do the bazillions of calculations required to train a neural network.
So suddenly nVidia is the biggest hardware manufacturer in the world, and still growing. Their hardware is incredibly fast, but consume vast amount of power to be so. A cabinet of nVidia servers the size of your refrigerator consumes about the same amount of power as 50 houses. A hyperscale data center with thousands of racks is the same as a small city.
And they use a lot of water. Instead of mechanical air conditioning like we have in our homes, and like what used to be the norm in data centers, modern large data centers are almost all cooled by evaporating water. A massive water line was run to the AWS Britton Rd campus. I imagine the lines of the same size or greater were run to Cosgray and Scioto Darby. There a new water plant being built on the Scioto River at Home Rd, in Delaware County. How much of that capacity will be used to make up for the water AWS, Google, Meta, Intel, Microsoft et al will be consuming?
And here's the thing. Remember the algorithms I mentioned above? There are PhDs working real hard to find a better approach to training neural networks. I would not be the least bit surprised if there in an "aha moment" soon which shows a way to do that training with a fraction of the resources. Then what happens?
There is already a pull back in the pace of hyperscale data center build out. Folks got a little spooked by the announcement of a new Chinese algorithm which is much more efficient. What happens if, after local utilities and governments sink $billions into upgrading their infrastructure to support their new hyperscale neighbors, they shut a bunch of them down, walk away, and stop paying their utility bills.
AEP is wise to this risk, and they want the PUCO to allow them to create a tariff (rate structure) that forces hyperscalers who cause them to make $billions of infrastructure investments (remember those two huge transformers hauled through the middle of Hilliard last summer?) to commit to 20 or 30 year minimum payments - to ensure that the hyperscalers pay for all that stuff, not the consumers left behind. The hyperscalers are of course fighting this at the Statehouse. And they have a LOT of money to do so.
My last professional role before retiring in 2000 was as the VP/General Manager of the Global Web Hosting business for UUNET, then a division of MCI Worldcom. I cut my teeth in the Computer Technology division of CompuServe, which was a group which built our three data centers (Upper Arlington, Dublin and Hilliard) and operating the large number of mainframe computers on which we ran the products used by our customers. The data center we built at 5000 Britton Pkwy (now the Verizon building) was one of the largest in the region. Today, it's not much larger than the power management rooms in the hyperscale data center buildings Amazon Web Services has built next door, each building 10x the size of our data center - and there are 5 of them on this site. And 5 more in Dublin. And 5 more in New Albany. Another batch on Cosgray Rd are nearly finished, and yet another on Scioto Darby is well along.
We picked the site on Britton Rd for exactly the same reasons AWS did: Access to plentiful electric power (we connected to two substations for redundancy), easy network connectivity (again, to two separate Ohio Bell Central Offices, plus our own private fiberoptic loop between the three data centers), cheap land, and a friendly local government.
I loved it when the first datacenters started appearing in Hilliard. AWS worked a sweet deal with the school district to pay, in lieu of property taxes, somewhere around $1.25-$1.50 per square foot annually to the school district (Norwich Township got left out of that deal of course). This will become a massive revenue source for Hilliard Schools, and should be immune to the threat of the elimination of property taxes (unless AWS anticipated that possibility and has an escape clause in their revenue agreement). Best of all, those data centers would put little demand on local resources: not schools or the police or the fire department. No traffic on the roads to speak of when construction is complete. No pollution other than the rare occasions when they have to run their diesel generators (and they have a LOT of them).
But it's getting out of hand - here an nationwide. The drive to built more and more inconceivably large neural networks to support nano-targeted marketing and a myriad of commercial and industrial applications has led to a whole real estate sector focused on finding greenfields anywhere in the country (planet) capable of supporting data center campuses which consume as much electricity and water as a decent sized city. Check out an outfit called "Data Center Hawk" as one example.
This frenzy sits upon on a well-known set of algorithms for training neural networks that were first developed decades ago. But these algorithms require an intensity of calculations that were not economically or practically feasible given the computing systems of the time. But then someone figured out that the graphics cards - GPUs - being produced by folks like nVidia to enable more and more realistic experiences for the gamer universe were exactly the architecture needed to do the bazillions of calculations required to train a neural network.
So suddenly nVidia is the biggest hardware manufacturer in the world, and still growing. Their hardware is incredibly fast, but consume vast amount of power to be so. A cabinet of nVidia servers the size of your refrigerator consumes about the same amount of power as 50 houses. A hyperscale data center with thousands of racks is the same as a small city.
And they use a lot of water. Instead of mechanical air conditioning like we have in our homes, and like what used to be the norm in data centers, modern large data centers are almost all cooled by evaporating water. A massive water line was run to the AWS Britton Rd campus. I imagine the lines of the same size or greater were run to Cosgray and Scioto Darby. There a new water plant being built on the Scioto River at Home Rd, in Delaware County. How much of that capacity will be used to make up for the water AWS, Google, Meta, Intel, Microsoft et al will be consuming?
And here's the thing. Remember the algorithms I mentioned above? There are PhDs working real hard to find a better approach to training neural networks. I would not be the least bit surprised if there in an "aha moment" soon which shows a way to do that training with a fraction of the resources. Then what happens?
There is already a pull back in the pace of hyperscale data center build out. Folks got a little spooked by the announcement of a new Chinese algorithm which is much more efficient. What happens if, after local utilities and governments sink $billions into upgrading their infrastructure to support their new hyperscale neighbors, they shut a bunch of them down, walk away, and stop paying their utility bills.
AEP is wise to this risk, and they want the PUCO to allow them to create a tariff (rate structure) that forces hyperscalers who cause them to make $billions of infrastructure investments (remember those two huge transformers hauled through the middle of Hilliard last summer?) to commit to 20 or 30 year minimum payments - to ensure that the hyperscalers pay for all that stuff, not the consumers left behind. The hyperscalers are of course fighting this at the Statehouse. And they have a LOT of money to do so.