Scott has been sending me some good stuff on data centers — let’s take a look. First, Maine is about to ban the construction of new, big data centers until November, 2027:
Maine has some of the country’s highest residential electricity prices, and elected officials are concerned that a surge in data-center power demand might further inflate costs. The AI build-out is driving up electricity costs for consumers in some parts of the country, and at the same time generating large tax revenues for local governments that continue to court developers.
[…]
“I think Maine is the canary in the coal mine,” said Anirban Basu, chief economist for the Associated Builders and Contractors, a construction trade group that counts members who work on data centers. “Maine will be the first of many states to have such moratoria.”
Legislators have introduced measures to temporarily ban or restrict data centers in New York, South Carolina, Oklahoma and other states. In Ohio, one of the top states for data-center development, a group of rural activists is collecting signatures to put a statewide ban of large data centers on a November ballot.
Many other municipalities and counties, especially small ones in Michigan and Indiana, already have imposed their own temporary pauses. Denver and Detroit are among major cities considering such bans.
Data-center developers are growing increasingly wary of community and political opposition as they hunt for powered land across the country. Proposed local laws restricting data centers are “a red flag,” said Tracey Hyatt Bosman, a site selection consultant at BLS & Co. who works with data-center developers. “They do limit where we are looking,” she said.
Scott also highlighted this election result:
FESTUS, Mo. – Major changes are coming to the Festus City Council after all four incumbents on the ballot were defeated in Tuesday’s municipal election.
Election officials said the voter turnout in Festus Tuesday was higher than other places in Jefferson County. A controversial data center plan recently approved by the Festus City Council was a motivator for some voters heading to the polls. The council approved the project by a 6-2 vote on March 30.
[…]
The result was that all four Festus councilmembers seeking reelection lost their races to challengers by wide margins. Three of the four incumbents voted yes in March on the data center plan.
Not everybody opposes data centers — just up the road from where I’m sitting, Laramie County (Cheyenne) Wyoming has approved a 2.7 gigawatt data center which will consume three times the power consumed by the entire state of Wyoming. That’s a jaw-dropping amount of power.
Honest polling on data centers (i.e., not the polling done by hack David Shor) shows that they have a negative effect on the environment, home costs and quality of life. (That’s a Pew poll.) Even a Politico poll, which I have to assume was paid for by donors, shows that people won’t tolerate more than a $25/month increase in their electric bills as a result of data centers.
As a Computer Science guy, I would be very surprised if AI can’t be made far more efficient. The history of Computer Science is full of optimizations that can yield order of magnitude improvements in efficiency. Plus, the race to lock people into an AI provider means that a ton of AI “cycles” (work) are being wasted by, for example, Google grinding through all my mail and providing AI suggestions that I ignore. When people actually pay for AI, they’re going to be far more choosy about what they spend their money on.
AI companies are spectacularly unprofitable:
Spending on AI training will be staggering. In 2028, OpenAI projects spending $121 billion on computing power for its AI research. The estimate for 2029 is slightly higher, before AI model training costs dip back below $100 million in 2030. (This year, for perspective, the company expects to spend just over $25 billion on AI model training.)
Anthropic’s totals are smaller but still climb steadily, surpassing $30 billion in 2029.
These losses come despite an expected surge in revenue at both companies. OpenAI’s revenue is projected to nearly double annually, reaching roughly $275 billion in 2030. Anthropic expects to approach $150 billion in 2029.
All of the AI companies are betting that they will have paying customers, but at this moment, I wouldn’t take that bet. What are people really going to be willing to pay for AI? Will Google bundle in enough ads with Gemini (their AI) to let it be “free” to everyone, just as they have with Gmail, Voice and their other free services?
The tech companies have all gotten into bed with Republicans. Datacenters are, on the whole, unpopular. Democrats should be opposing these monstrosities, especially because it’s almost certain that way, way too many of them will be built.
