What 10,000 US Council Comments Mean for Data Center Buyers
Short answer (checked 2026-11-02): American Sentiment, a project of the San Francisco start-up Prophecy Gov, has coded 10,202 public comments about data centers made at 417 US local governments in 2026. 90.5% of the speakers were opposed, and at 73% of the 898 meetings nobody spoke in favour. We downloaded the coded data and recounted the concerns ourselves: water comes first (35.8% of comments), grid capacity is at 25.7%, noise at 24.5%. If you rent cloud or dedicated servers this is not a US curiosity. Every item on that list is a cost or a delay for whoever builds capacity, and Europe now makes operators publish the numbers behind most of them.
Source. "America on Data Centers" at americansentiment.com, discussed on Hacker News (31 points and 99 comments on 2026-11-02). The site itself warns that public comments do not always match what the public thinks: the people who come to a council meeting are mostly the people who object.
What the dataset says
The site lists each comment with the speaker's stance and the concerns raised. Its own headline figures, as published on 2026-11-02:
| Stance | Comments | Share |
|---|---|---|
| Opposed | 9,232 | 90.5% |
| In favour | 718 | 7.0% |
| Asked questions | 252 | 2.5% |
Of the 898 meetings, 658 had no one speaking in favour. The only group that mostly supported a project was union workers (79% in favour, 183 comments), and 88% of comments raised more than one concern. Lack of transparency appears in 76% of meetings, more often than any physical concern: people object to how projects are approved almost as much as to what is built.
Recount it yourself: dc_concerns.js
The site's Explore tab loads one public file, data/coding.json: a list of the 43 concern codes and, per comment, the indexes of the concerns it raised. This script counts them. Save it as dc_concerns.js:
// dc_concerns.js - count coded concerns in American Sentiment's data/coding.json
// usage: bun dc_concerns.js coding.json [top]
const [file, top = 8] = process.argv.slice(2)
const { vocab, rows } = JSON.parse(await Bun.file(file).text())
const packs = Object.values(rows) // [subject, relationship[], concerns[], benefits[], requests[], score]
const n = packs.length
const count = new Map()
let multi = 0
for (const p of packs) {
if (p[2].length > 1) multi++
for (const i of p[2]) count.set(i, (count.get(i) || 0) + 1)
}
const pct = c => (100 * c / n).toFixed(1).padStart(5) + '%'
const line = (c, name) => console.log(String(c).padStart(5), pct(c), name)
console.log(`coded comments: ${n}, with 2+ concerns: ${(100 * multi / n).toFixed(1)}%`)
;[...count].sort((a, b) => b[1] - a[1]).slice(0, +top).forEach(([i, c]) => line(c, vocab.concerns[i]))
// the concerns a hosting provider has to answer too
console.log('-- infrastructure concerns')
for (const k of ['water_consumption_supply', 'electricity_demand_grid_capacity_reliability',
'noise_vibration', 'utility_rate_increases_cost_shifting', 'onsite_backup_power_fuel_risk'])
line(count.get(vocab.concerns.indexOf(k)) || 0, k)
curl -sLo coding.json https://www.americansentiment.com/data/coding.json
bun dc_concerns.js coding.json 8
Output, run on a DCXV Cloud server with Bun 1.4.4 on 2026-11-02:
coded comments: 10133, with 2+ concerns: 88.5%
3628 35.8% water_consumption_supply
3363 33.2% health_quality_of_life
3357 33.1% public_notice_transparency_participation
2862 28.2% company_trust_conflicts
2605 25.7% electricity_demand_grid_capacity_reliability
2509 24.8% weak_rules_enforcement_accountability
2482 24.5% noise_vibration
2276 22.5% rushed_process_inadequate_study
-- infrastructure concerns
3628 35.8% water_consumption_supply
2605 25.7% electricity_demand_grid_capacity_reliability
2482 24.5% noise_vibration
1663 16.4% utility_rate_increases_cost_shifting
765 7.5% onsite_backup_power_fuel_risk
The file holds 10,133 coded comments, 69 fewer than the headline 10,202. We do not know which ones are missing, so our shares can differ from the site's by a fraction of a point (water: 35.8% here, 36% on the site). The ranking is the same.
Five concerns, five questions for your provider
Residents and customers worry about the same physical facts from opposite sides. A town asks whether the building will drain its water or its grid; a customer should ask whether the building can keep running and grow. Ask the provider who hosts your servers:
| Concern (share of comments) | Why it matters to a customer | Question to ask |
|---|---|---|
| Water use (35.8%) | Evaporative cooling needs water permits that can be capped | What cooling does the site use, and what is its WUE? |
| Grid capacity (25.7%) | No new power means no new racks for your growth | Is there contracted power for more servers in this facility? |
| Noise (24.5%) | Noise limits can block chillers and generators, and expansion with them | Is the building permitted for its current and planned load? |
| Utility rates (16.4%) | Energy costs reach customers as surcharges | Is power on a fixed contract, and how are energy changes passed on? |
| Backup fuel (7.5%) | Generators are the last line during a grid failure | How many hours of fuel are on site, and when was the last full-load test? |
Europe turns the answers into data
In the US the fight happens at council meetings. In the EU, more of it now runs through grid rules and mandatory disclosure:
- Reporting. Under Article 12 of the Energy Efficiency Directive and Delegated Regulation (EU) 2024/1364, every data centre with an installed IT power demand of at least 500 kW reports its energy use, water use, waste heat and grid participation to a European database each year. The report for 2025 was due by 15 May 2026 (Swedish Energy Agency guidance, checked 2026-11-02).
- Label. On 21 September 2026 the European Commission adopted a delegated regulation for a mandatory A to G sustainability label for the same 500 kW sites, graded on PUE and WUE. Parliament and Council have a scrutiny period first, and the first labels are expected in 2027 (Linklaters, 2026-10-14).
- Grid. Ireland's regulator CRU (decision CRU2025236, December 2025) now requires a new data centre to bring generation or storage matching its maximum import capacity, on site or nearby, and to match at least 80% of its yearly demand with new renewable projects in Ireland within a six-year glide path.
For a buyer this is good news: from 2027 the answers to the first two rows of the table should be printed on a label instead of hidden in a sales call.
What it means if you rent servers
- New capacity gets slower; existing capacity is what you can actually buy. A campus that needs a new grid connection and a council vote can take years. Racks in a building that already has power, water and permits are available now. That is why we run in two existing Tier III facilities, compared in our guide to Tier III data centres in Europe.
- Do not pay for headroom you will not use. Power and hardware are both getting dearer: server memory alone rose sharply this year, see server RAM in 2026. Size by measured use, as in our cloud server sizing guide, and grow when the numbers say so.
- Stay portable. A site that cannot expand is only a problem if you cannot move: keep images, backups and DNS in a form you can restore elsewhere in hours, not weeks.
Ask us the same questions
Our cloud servers and dedicated servers run in Tier III data centres in Prague and Covilha. Send the five questions from the table to sales@dcxv.com and we will answer them for both sites, with the facility data we can share.
