Fun with Data - Data Center Energy Consumption
There is a great deal of interest in the power requirements of data centers as the growth of same has continued to flourish across the globe. AI model developers are requiring an ever increasing amount of computing power to train and fine-tune models, and that calls for a rapid growth in data center capacity. By next year it is estimated that 40% of global data center power consumption will come from AI workloads and that consumption is expected to continue to grow. Major data center build projects that have been announced are large (see below) and usually put into service in phases, but timing has always been an iffy metric for such projects in the past. Now that the AI boom has pushed the needs for data center capacity, we expect most near-term projects will open as quickly as possible, while those preliminarily scheduled for the end of the decade will take a more measured approach. That said, utilities and the local, state, and county-wide power grids must prepare in advance to meet this growing need, and that is not an easy task given the regulatory issues in many countries, including the US.
In the US energy grid consumption breaks down as follows, with the data center share included in the ‘commercial’ category. Given the penchant for data center growth and energy demand, we expect the commercial category to surpass residential share in 2026 or 2027.
This is certainly not only a US problem, although we are the largest consumer of energy for data centers in the world (The US represents 45% of global data center energy consumption – China is #2 @25%), There are a number of countries who are trying to capitalize on the data center boom but are finding that they cannot guarantee adequate power for same, as local or country grids are already over-taxed. Here’s an example:
Spain’s government has a goal of making the country a hub for European data centers, but at the end of April the entire country fell into a several hour blackout, causing the Spanish government to blame Iberdrola and Endesa (Spanish energy utilities) for the outage, while the utilities blamed the government. Underlying the controversy was the fact that about 50% of the requests to access the country’s grid in 2024 were rejected, with roughly 1/3 of those requests coming from data centers. It seems that more than 83% of grid connection points in the country’s electricity distribution grid cannot process new connection links because of grid saturation, slowing plans for the country’s data center hub dream and straining an already strained grid.
One of the issues that surrounded this problem is the rate of return the government guarantees for investors in new electricity capacity and infrastructure, the TRF (Tasa de Retribucion Financiera or Financial Return Rate). Too low a rate and investors look for higher returns in other countries. Too high a rate and taxpayers, who are footing the bill, protest, so it seems that the Spanish government is considering ‘rebalancing’ that rate to make it more attractive for investors. The government wants to increase the rate from 5.58% to 6.46% and is concerned that too high a high rate will create a data center ‘bubble’, while the utilities say the new rate is not high enough to keep investors from shifting to more lucrative countries. This is indicative of only one aspect of the real ‘red tape’ that holds back electrical energy capacity growth, and as data center projects continue to get funding, the grids in which they operate less so. This increases the stress on residential, commercial, and industrial customers also.
Most of the data center projects we see are phased, in that they can be built out over time as the AI market demand grows. However whether the lack of capacity is government bureaucracy or an aging grid, data center energy demand is increasing. The numbers might seem small relative to consumer demand or even commercial demand in the table below, which shows those countries where data center energy demand is a large part of total consumption and how it has changed from just 2 years ago, but even small increments to demand can disrupt a grid’s integrity, and that does little to attract new data center investment. Planning for capacity expansion is also a very difficult job in light of the sudden influx of data center demand over the last few years, and both governments and utilities are very wary over the accuracy of data center demand forecasts, given its recent rapid rise. If AI demand weakens or data scientist figures out how to train with a more efficient process (DeepSeek) we could wind up with lots of excess energy capacity that will take many years to absorb.
Spain’s government has a goal of making the country a hub for European data centers, but at the end of April the entire country fell into a several hour blackout, causing the Spanish government to blame Iberdrola and Endesa (Spanish energy utilities) for the outage, while the utilities blamed the government. Underlying the controversy was the fact that about 50% of the requests to access the country’s grid in 2024 were rejected, with roughly 1/3 of those requests coming from data centers. It seems that more than 83% of grid connection points in the country’s electricity distribution grid cannot process new connection links because of grid saturation, slowing plans for the country’s data center hub dream and straining an already strained grid.
One of the issues that surrounded this problem is the rate of return the government guarantees for investors in new electricity capacity and infrastructure, the TRF (Tasa de Retribucion Financiera or Financial Return Rate). Too low a rate and investors look for higher returns in other countries. Too high a rate and taxpayers, who are footing the bill, protest, so it seems that the Spanish government is considering ‘rebalancing’ that rate to make it more attractive for investors. The government wants to increase the rate from 5.58% to 6.46% and is concerned that too high a high rate will create a data center ‘bubble’, while the utilities say the new rate is not high enough to keep investors from shifting to more lucrative countries. This is indicative of only one aspect of the real ‘red tape’ that holds back electrical energy capacity growth, and as data center projects continue to get funding, the grids in which they operate less so. This increases the stress on residential, commercial, and industrial customers also.
Most of the data center projects we see are phased, in that they can be built out over time as the AI market demand grows. However whether the lack of capacity is government bureaucracy or an aging grid, data center energy demand is increasing. The numbers might seem small relative to consumer demand or even commercial demand in the table below, which shows those countries where data center energy demand is a large part of total consumption and how it has changed from just 2 years ago, but even small increments to demand can disrupt a grid’s integrity, and that does little to attract new data center investment. Planning for capacity expansion is also a very difficult job in light of the sudden influx of data center demand over the last few years, and both governments and utilities are very wary over the accuracy of data center demand forecasts, given its recent rapid rise. If AI demand weakens or data scientist figures out how to train with a more efficient process (DeepSeek) we could wind up with lots of excess energy capacity that will take many years to absorb.
Data Summary
- 35.7% of countries shown saw data center share over 5%
- 71.4% of countries shown saw data center share equal or greater tha 2.5%
- 98.6% of countries shown saw data center share growth of 20% or more over the 2 year period.
- 78.6% f countries shown saw data center share growth of 30% or more over the 2 year period.
RSS Feed