Community Discussion · Tracks

Data Center Backlash Is More Than Just Noise Pollution

PM YuanSep 22026/09/02 33 views

Why does almost everyone working in AI start hating the infrastructure behind them?

Data centers are hated. Noise and ugliness are mentioned, but a more critical reason is that this wave of AI expansion has spread the costs to communities, while benefits are seen by only a few.

Previously, when I worked on AI imaging at United Imaging, I spent every day with radiologists. Later, I noticed a pattern: whether technology lands depends not just on model accuracy, but equally on who bears the hassle and who gets the benefit. Doctors are willing to use AI because it saves them writing one report or missing one small lesion; hospitals are willing to purchase because the process works. If frontline staff are forced to do more verification and take more responsibility, even advanced tech will be dragged along reluctantly.

Data centers are currently stuck in this position too.

External reasons for hating it are practical: noise, land occupation, water consumption, pushing up local electricity prices, and fears of jobs being eaten by automation. Behind these complaints is cost of living. MIT Technology Review had a quote roughly meaning this: nobody wants to see their electricity bill go up.

But inside the AI circle, looking at it is a different narrative. Compute power is the new means of production; training needs to run, inference needs to run, models need continuous updates.

The problem is, ordinary users don't perceive this. Nate Silver has a poignant saying.

Only those already in AI see the opportunities; outsiders just see a chatbot that gets slightly smarter every year.

This progress is life-or-death for product people, but for ordinary people, it just seems slightly cleverer. This gap is the source of the hate.

I think the bigger trouble is that the product logic of data centers has always lacked a key user: the neighbor. Previously, everyone assumed it was enterprise infrastructure; as long as the government approved it, electricity was connected, and water was sufficient, it was considered landed. But AI pushed it from the backend to the frontend. People next to a server room are forced to become users. They didn't buy the service, yet they bear the noise, construction, electricity usage, and subsidy disputes first.

Medical AI has similar pitfalls. When we evaluated an imaging model before, beyond paper metrics, we asked: Has clinical validation been done? What is the actual feedback from doctors' usage? Who signs off on misjudgments? Does anyone pay for the extra clicks added to the process? Data centers should learn this lesson too: Have local grid pressures been calculated clearly? Where does the water come from? Can benefits flow back to the community? How much disturbance do construction and operations cause to residents?

This is a business problem. If a product causes significant negative experiences for a large number of non-users, the fiercer the expansion, the bigger the backlash. Regulation will come, site selection will become more expensive, and financing will shift from storytelling to risk analysis. The more popular AI applications become, the harder it is to hide these costs.

Looking ahead, I predict competition among data centers will slowly shift from who has more cards and cheaper electricity, to who can hide themselves within the community. Low noise, water recycling, shared grids, local employment, community revenue sharing—these sound like PR buzzwords, but they will become hard conditions for site selection. We might even see smaller inference nodes closer to users, splitting some load away from giant server rooms.

Going forward, watch whether data centers can account for noise, water, electricity, and revenue to the community. If they can't calculate it clearly, the hate will continue.

📌 This article is compiled from Guardian Tech, original link https://www.theguardian.com/uk-news/video/2026/sep/01/why-does-everyone-hate-datacentres

Tracks

2 replies