from Birch Gold Group:
In recent months I’ve been asked by numerous readers to give my take on the AI data center issue and the debate it has ignited across the U.S. In response, I’ve jumped into an interesting and steep learning curve, but it seems to parallel information I’ve covered in previous articles on AI.
From what I can tell, the politics of the data center debate don’t break cleanly along party lines. There are Democrats and Republicans on both sides, though their reasons for opposing the centers are not necessarily the same. The Trump Administration has firmly embraced accelerated AI and data center development, framing it as an issue of economic competitiveness and national security. Even so, some of the strongest resistance is coming from deeply conservative communities.
TRUTH LIVES on at https://sgtreport.tv/
I call it “infrastructure,” but I have to say, one of my biggest apprehensions about data centers is the mismatch between their enormous physical footprint and the benefits that flow directly to local communities. They certainly create construction work, permanent jobs and tax revenue. But these are extraordinarily capital- and energy-intensive facilities, and I’m struggling to see why the benefits should automatically outweigh the local costs.
I’ll try to break it down in basic terms…
What do data centers actually do?
Much of the current boom in new hyperscale data centers is being driven by AI. These facilities provide the enormous computing capacity required for two basic jobs: Training AI models, and running those models once they’re deployed, a process called “inference.”
The most ambitious AI campuses contain enormous clusters of specialized chips used to train so-called “frontier” models. Training is computationally intensive and can run for long periods at very high utilization, which is one reason these facilities consume so much electricity. Inference – actually serving all those AI requests after a model has been trained – requires substantial computing capacity of its own.
And the computing equipment doesn’t become obsolete once a single model is finished. The same infrastructure can be used for additional training runs, new generations of models and the growing workload required to serve existing AI systems.
We can see and use cars, we can see and use televisions and refrigerators, we can see the workers on the factory floor and we can see the jobs these products create. Not so much with AI and data centers, and this is one of the problems with trying to convince the public that such operations are necessary for the future well-being of the country.
In some cases AI training facilities convert partially or fully to inference operations. Meaning they handle the day-to-day functions of existing AI. So, every time you ask ChatGPT a question, or ask it to create an image for you, or ask it to make a ridiculous, uncanny valley video for you, that request is processed by a data center somewhere.
All of that computing uses electricity – a lot of it. A 2026 update from Lawrence Berkeley National Laboratory estimates that data centers could consume about 11.8% of all U.S. electricity by 2030, with plausible scenarios ranging from 9.5% to 15.3%. AI isn’t solely responsible for that demand, but it is one of the major forces driving the increase.
If this all seems highly centralized, that’s because it is. There are also legitimate privacy questions surrounding cloud-based AI. Depending on the service and your settings, conversations may be retained and some may be used to improve future models. Both ChatGPT and Gemini now offer data-use controls that limit how conversations are used. So maybe every interaction doesn’t automatically becomes training material. Still, using a remotely hosted AI service necessarily means entrusting that information to someone else’s infrastructure, security and internal policies.
