I think you can ignore onsite solar and wind as providing an infinitesimally small share of the power required by a data center, and nuclear sources are years away, if they come at all. If you're in an area with plenty of cheap natural gas, this might work. In other places, however, it might not be cost competitive with grid power. Furthermore, in populated areas self generation heightens concerns about air pollution and noise.
In addition, there are relatively unknown heat island effects of data centers. Essentially all of the power going into a data center turns into heat, which at the size of some of these hyperscale campuses is enough to affect local weather. If you're generating power onsite, the problem is compounded. If you use open-cycle gas turbines, with a thermodynamic efficiency of less than 25%, heat output rises four-fold or more. Using combined-cycle gas turbines is more complex, but you might get efficiencies greater than 50% - which still doubles the heat output.
The bigger unknown here is how much power is needed for computation. Recent published research from the University of Minnesota claims to have found a way to reduce the basic power for computation by a factor of 1000 or more. I have no idea if this is feasible, but the research group seems credible and is backed by DARPA and Cisco. If they can achieve even a fraction of the claim, it turns the whole data center model on its head.
Lynne, thanks for this piece. Question: have you come across any information about data center operators eyeing the Permian Basin as an ideal place to build new AI data centers due to low cost energy in natural gas, an unregulated electricity grid, water for cooling and unpopulated land?
What are the plausible energy sources for plausible data center micrpgrids? If we are a talking 100's of MW, theybare not going to have the space for any appreciable fraction being wind or solar without going through the main grid anyway.
Getting down to that *small* of scale with nuclear would need a sea change in regulation.
I think you can ignore onsite solar and wind as providing an infinitesimally small share of the power required by a data center, and nuclear sources are years away, if they come at all. If you're in an area with plenty of cheap natural gas, this might work. In other places, however, it might not be cost competitive with grid power. Furthermore, in populated areas self generation heightens concerns about air pollution and noise.
In addition, there are relatively unknown heat island effects of data centers. Essentially all of the power going into a data center turns into heat, which at the size of some of these hyperscale campuses is enough to affect local weather. If you're generating power onsite, the problem is compounded. If you use open-cycle gas turbines, with a thermodynamic efficiency of less than 25%, heat output rises four-fold or more. Using combined-cycle gas turbines is more complex, but you might get efficiencies greater than 50% - which still doubles the heat output.
The bigger unknown here is how much power is needed for computation. Recent published research from the University of Minnesota claims to have found a way to reduce the basic power for computation by a factor of 1000 or more. I have no idea if this is feasible, but the research group seems credible and is backed by DARPA and Cisco. If they can achieve even a fraction of the claim, it turns the whole data center model on its head.
Lynne, thanks for this piece. Question: have you come across any information about data center operators eyeing the Permian Basin as an ideal place to build new AI data centers due to low cost energy in natural gas, an unregulated electricity grid, water for cooling and unpopulated land?
What are the plausible energy sources for plausible data center micrpgrids? If we are a talking 100's of MW, theybare not going to have the space for any appreciable fraction being wind or solar without going through the main grid anyway.
Getting down to that *small* of scale with nuclear would need a sea change in regulation.
What's left other than something natural gas?