October 21, 2024
By Debbie Coffey, AI Endgame
Hurricanes Helene and Milton knocked out the power for millions of people in North Carolina, Georgia and the Florida. For the fortunate among us not impacted by the storms and power outages, we were witnessing yet another devastating weather event, and wondering “are we next?”
People living in the Southwest are seeing summer temperatures rise as high as 122 degrees (and that’s not even in Death Valley).
In 2021, a winter storm in Texas knocked out power for millions of people for days. One person described the temperature inside of his house as just above the freezing mark and said the family goldfish froze in its tank. [1]
Extreme weather is causing most of us to realize how important power is to our survival.
I recently did some research on the U.S. power grid, and I learned “the grid” actually consists of three big grids (the Eastern and Western grids and the independent Texas grid, and about a dozen smaller regional grid systems).
I also found out our power grid is aging and struggling to even meet current demands.
The AI industry would like to keep under wraps the fact that datacenters are consuming huge and ever-growing amounts of power. (So, naturally, I’d like to talk about that.)
AI DATACENTERS
Datacenters are massive facilities with thousands of computer servers running non-stop, 24/7.
Datacenters require power for servers (which draw most of the power), cooling systems, storage systems, networking equipment, backup systems, security systems, and lighting.
Large ‘hyperscale’ datacenters, range in size from 100,000 square feet to several millions of square feet, with tens of thousands of servers, and can use over 100 megawatts of power. [2] To put this into context, 100 megawatts of power could power around 80,000 homes (per U.S. Dept. of Energy, 2020). [3]
There are many reasons datacenters are now using more electricity:
· Phone data backed up to the “cloud” is stored in a datacenter. Datacenters also store data including social media, financial transactions, and government operations. [4]
· A Chat GPT request consumes about 10 times more electricity than a Google search, according to the International Energy Agency. [5]
· Training AI models, involves scraping massive volumes of data from social media, the internet and transcribed YouTube videos, then processing this data in units performing rapid mathematical calculations that run continually for thousands of hours. This requires gigawatts of electricity on a scale that has never been required before. [6]
· Competitive tech companies are racing to build more powerful frontier models (“frontier” models are AI models that exceed the capabilities of present AI models). [7] The bigger the AI model, the more computation required, and frontier models are gigantic.
AI is expected to cause a 160% surge in power demand from datacenters by 2030. [8]
Microsoft and OpenAI are planning to build a massive $100 billion datacenter and supercomputer (dubbed “Stargate”) in the U.S. by 2028. [9]
Increases in costs per megawatt
As AI demands on the electric grid are mounting with the increased use of smart technologies, many of the generators in the US grid remain “on” around the clock. Most energy used now vacillates during the day and during the year, and costs about $30 per megawatt.
However, when electric demand, including additional power consumed by datacenters, exceeds the supply of these base power plants, utility companies turn on peaker plants that can ramp up quickly but are more expensive (about $1,000 per megawatt). Peaker plants (peaking power plants) are power plants that generally run only when there is a high demand, known as peak demand, for electricity.
Your electric bill will go up
The use of these extra power companies means your electric bill will go up. Your bill could even double. (That really gives us something to look forward to, doesn’t it?)
In 10 years, as electrical demand increases, our power grid infrastructure won’t be able to keep up with the demand.
Electricity rationing
We’re likely facing electricity rationing. [10]
Are the datacenters going to cut back or shut down if electricity rationing is needed? I bet they’ll keep right on running at full capacity, while the rest of us pay the price.
Are people in the Southwest going to swelter without air conditioning when it’s 122 degrees in the summer? What if the gas pumps in your area aren’t pumping gas because there’s no electricity? (And, what if we can’t watch football?)
We already have brownouts (a partial reduction in power to sections of an electrical grid) and blackouts (an unplanned, total loss of power to part, or all, of an electrical grid).
How many datacenters are there?
There are about 8,000 datacenters globally.
The US has over 2,700 datacenters. We have more datacenters than any other country. [11]
“Edge” datacenters are being built close to major population centers (edge computing involves processing data closer to where it is generated). [12] There are many datacenters near big cities like Los Angeles, Dallas, Chicago, Atlanta, and New York City. Most datacenters in the U.S. are near Washington DC, in Northern Virginia, with nearly 300 datacenters in Loudon County alone (known as Data Center Alley). [13]
It was estimated electricity demands of datacenters around the world would more than double from 2022-2026 (in only 4 years).
The climate pledges that major tech companies set for themselves are going by the wayside because of the huge amounts of energy required by AI. Even if these tech companies talk up a good game about some improvements or changes, just remember that the energy needed for AI will be ever-increasing.
A “black hole” for energy
The power needed for the ever-increasing demands of AI datacenters will create a “black hole” for future energy. While the demand for energy doubled in only four years, this will increase in coming years as more, and larger, datacenters are built.
The Biden Administration started the ball rolling to work on solutions by adding grid enhancing technologies to our existing infrastructure, but this process involves individual states and will take years.
Some other possible solutions to increase energy include combating inefficiencies, improving chip hardware, building transmission lines that can export power from one part of the country to another, and creating more wind, solar and nuclear energy.
Next week: The push for nuclear power (as if we haven’t learned our lessons from Chernobyl and Fukushima).
Please share what you learn with others. We need more public awareness of all AI issues so that we can work together to have power over our future.
What you can do:
1) Call your representatives and tell them you “want regulations to pause AI now, until strong AI safety laws are enacted.”
Find out how to contact your Congressional representatives here:
https://www.house.gov/representatives/find-your-representative
Find out how to contact your Senators here:
https://www.senate.gov/senators/senators-contact.htm?Class=1
2) Support (and if you can, make donations) to organizations fighting for AI Safety:
Pause AI
Center for Humane Technology
https://www.humanetech.com/who-we-are
The Center for AI Safety
Read past AI Endgame newsletters and share links with your friends:
#1 - AI Endgame: Introduction Read HERE.
#2 - AI Endgame: Risk of Human Extinction & AI regulations Read HERE.
#3 - AI Endgame: Rogue AIs Read HERE.
#4 - AI Endgame: Political Deepfakes Read HERE.
#5 - AI Endgame: Nobel prize winner Geoffrey Hinton, “The Godfather of AI,” warns of AI dangers Read HERE.
I’ve been doing investigative journalism for 13 years and I hosted a BlogTalk radio show for 6 years.
In 2023, over 600 AI researchers, scientists and engineers warned that there is a great risk that AI could lead to human extinction. AI Endgame will provide you with information in an easy-to-understand format and alert you to actions you can take.
[1] https://www.cnn.com/2024/10/18/climate/us-electric-grid-extreme-weather-climate/index.html
[2] https://dgtlinfra.com/data-center-power/
[3] https://energyinnovation.org/2020/03/17/how-much-energy-do-data-centers-really-use/
[4] https://time.com/6987773/ai-data-centers-energy-usage-climate-change/
[5] https://www.cnn.com/2024/09/12/tech/ai-chatgpt-white-house-power-energy/index.html
[6] https://www.cnn.com/2024/07/05/opinions/artificial-intelligence-electricity-grid-demand-kuntz/index.html
[7] https://time.com/6987773/ai-data-centers-energy-usage-climate-change/
[8] https://www.cnn.com/2024/09/12/tech/ai-chatgpt-white-house-power-energy/index.html
[9] https://www.crn.com/news/data-center/microsoft-and-openai-plan-100b-ai-centric-data-center-report
[10] https://www.cnn.com/2024/07/05/opinions/artificial-intelligence-electricity-grid-demand-kuntz/index.html
[11] https://www.newscientist.com/article/2342490-how-much-water-do-data-centres-use-most-tech-companies-wont-say/
[12] https://aws.amazon.com/what-is/edge-computing/
[13] https://time.com/6987773/ai-data-centers-energy-usage-climate-change/