Data Centers Could More Than Double Their Energy Draw by 2035. On X, the Argument Is Already About Who Pays.

A Bell Labs account posted something blunt this week: AI is scaling fast, but so is its energy demand, and data center consumption could more than double by 2035. The post linked to a paper on reducing energy use across models, hardware, and training without sacrificing performance, and it got modest traction — a handful of retweets, a few likes. But the framing of that post, "smarter AI, not bigger AI," landed in a conversation that had already moved past the question of whether AI has an energy problem and arrived at something sharper: who is going to pay for it.

That pivot is visible in how the language has changed. A few weeks ago, posts in this space leaned on carbon footprints and climate pledges. Now "energy consumption" has become the dominant phrase, and the posts carrying it aren't from environmental reporters — they're from people angry about utility bills. One account on X put it with the kind of directness that gets shared: data centers shouldn't force regular people to pay higher bills for AI output, and if they're going to consume that much power, they should face higher rates and be held to environmental standards, including noise pollution. That post has two retweets and not much reach, but it captures a sentiment that is spreading faster than its engagement numbers suggest — a move from ambient climate anxiety to specific, localized grievance.

On Bluesky, the framing is more technical but no less alarmed. One post this week noted that large data centers can consume up to five million gallons of water per day for cooling — comparable to a city of 10,000 to 50,000 people — with AI accelerating that demand. News outlets, meanwhile, are running a different story: breakthroughs, circular economy models, robots recycling disk drives at Microsoft, NVIDIA cutting emissions while crushing earnings. The gap between those two conversations is not new, but it is widening. The optimistic institutional coverage and the skeptical grassroots posts are now barely responding to each other — they've sorted into separate audiences with separate premises.

The one genuinely interesting counterweight came from a post about World Models — the idea that AI embedded in robotics and sensors, operating in the physical world rather than data centers, might require noticeably less energy than the current infrastructure model. That argument, if it holds, would reframe the entire debate: the problem isn't AI's energy appetite, it's the specific architecture we're using to feed it. That trade-off has been visible for a while to anyone paying attention. But the people on X furious about their electricity bills aren't waiting for the architecture to change — they're already deciding that the cost is being pushed onto them, and that framing, once it takes hold, tends to become the one that reaches legislators first.