US electricity generation data from 2015 to 2025 shows data centers accounting for around 300 TWh by 2025 while AI-related usage remains near zero until a small 2024 uptick

This is the single most important chart.

If AI were driving prices, you'd see a cluster top-right. You don't.

States with huge load growth (VA, TX, NV, ND, IA) sit at ~0c change in 5y. States with massive price hikes (CA, NY, MA, CT) have basically NO load growth.

AI data center growth is not correlated with electricity price changes.

"but I live in one of those states and my bill went up" – in nominal dollars. in real terms, bills are down in the most datacenter-intense states.

it's inflation, not AI, which is driving up energy prices.

I sorted 50 states into data center intensity quintiles. States with the highest datacenter energy share (Q5) VA, TX, NV, IA, OR, AZ... collectively have the lowest and slowest rising mean residential energy prices

Biggest real price hikes since 2019: CA, RI, DC, CT, ME, NY, MD, MA. All blue. All low data-center exposure.

Falling real prices: mostly red, many with heavy data centers.

AI isn't (yet) the villain. Policymakers complaining about AI might be.

If not AI, then what?

Politics.

Blue states pay ~6c/kWh more than red states, and the gap is widening.

Denuclearization, blocked pipelines, rigid renewable mandates, NIMBYism, expensive union labor – these are stated policies, not market forces and not AI.

Full writeup and complete methodological information for every chart available here:

https://murmurationstwo.substack.com/p/ai-is-not-hiking-your-electricity

AI is also getting more efficient.

For a fixed unit of intelligence, energy per query has fallen ~200x in 3.5 years (conservative estimate)

Aggregate demand is still climbing, yes. (cc Jevons) But intelligence is getting cheaper faster than any commodity in modern history.

States with lots of datacenters have cheap power, and datacenters go to cheap states.

Blue states have costly power because they are pursuing climate transitions, have NIMBY dynamics, and don't build pipelines & deprecate nuclear.

Case study: Texas

ERCOT's large flexible load capacity went from ~1.2 GW (2018) to ~9.5 GW (2025) – a gargantuan AI + crypto buildout.

Real residential prices barely moved. Cheap natgas, huge solar/wind/battery additions, and behind-the-meter campuses absorbed almost all of it.

@FiftyOne_50_ @ylecun this is AI slop

Now zoom into the top 10 AI datacenter states.

Real rates ARE ticking up in 2025 – especially in PJM (OH, VA). That's where the buildout is finally starting to bite.

But the magnitude is small, and most of these states still sit well below the US average.

@nic_carter On a related note that ppl get worked up about, I wish more people would see how much water is being used on golf courses vs. data centers daily

I'm talking about current data.

no one knows what's going to happen in the future. it's definitely the case that people blaming AI for bills now are wrong; these bills are the result of inflation and busted climate transition in blue states.

some states where massive buildouts have already happened, like Texas, have not seen concomitant energy hikes.

I specifically acknowledge the Maryland PJM issue in my full writeup: https://murmurationstwo.substack.com/p/ai-is-not-hiking-your-electricity

regardless of what happens _later_, it's useful to couch the discussion in current facts.

@nic_carter More generally, demand isn’t the problem. It’s transmission. Ever-increasing % of the bill, while generation is a shrinking %. Generation is getting cheaper all the time.

Smartly-located, flexibly-operated centers, especially those that bring their own power, actually help.

@fityourframe It’s just the current reality.

Will AI bite in the future? Maybe. But are policymakers especially in blue states wrong to blame AI today? Absolutely. They are wrong. Their own policies made power expensive.

@nic_carter So just to be clear, these things are not going to hike my electricity bills. They're not going to cause droughts. They're not going to cause noise or actual pollution. And they're not going to surveil me. Cool! Now let me get all of this in writing signed by the company involved

@nic_carter most havent been built yet, people are talking about new ones, your thread is irrelevant. We are already stuck with infrastructure bills. For ex my state: "Maryland citizens slapped with $2 billion power grid upgrade bill". You propagandists always lie.

https://www.tomshardware.com/tech-industry/artificial-intelligence/maryland-citizens-slapped-with-usd2-billion-grid-upgrade-bill-for-out-of-state-ai-data-centers-state-complains-to-federal-energy-regulators-says-additional-cost-breaks-ratepayer-protection-pledge-promises

The data backs the current snapshot — but the wave hasn't fully landed. ERCOT alone is forecasting 50+ GW of new data center load by 2030 with interconnection queues already 4–7 years deep. The next leg of price hikes will track that pipeline, and the right response is building generation and transmission now, not blaming AI for a decade of underinvestment.

@thetrocro now whether the AI people can flexibilize their loads like those exemplary bitcoin miners....

that would be a huge deal if someone could pull it off.