Stopping AI research and being against datacenters?

Every few months the word “pause” drifts back into the AI conversation, as if a handbrake could be gently lifted on a runaway technology. Elon Musk famously signed the Future of Life Institute’s 2023 open letter urging a six‑month halt on training models more powerful than GPT‑4; Sam Altman told the US Senate that licensing and international coordination may require temporary stoppages at certain capability thresholds; Dario Amodei’s Anthropic has gone furthest in operationalising that idea with a “Responsible Scaling Policy” that contemplates pausing when risk evaluations trip red lines. The research community remains split: Nature and other journals have catalogued both the prudence of breathing space and the risk that a pause entrenches incumbents or simply drives work elsewhere. In practice, what we’ve had is not a pause but a politics of pacing—soft law commitments, model evaluations, and the UK’s Bletchley Declaration—all inadequate substitutes for a shared, enforceable theory of limits.

While leaders debate the speed of thinking machines, communities are arguing about the buildings that house them. Across Europe and the US, datacentres have become proxies for deeper anxieties about power, water, noise, and fairness: Dublin’s grid constraints and connection limits, Amsterdam’s earlier moratorium and stricter zoning, Frankfurt’s tightening on new sites, and Northern Virginia’s very public rows over “data center alley.” The objections are not imaginary; the IEA’s 2024 analysis projects global electricity demand from data centres, AI and crypto could reach 4–6% of total consumption by 2026, and Ireland’s statistics agency estimates data centres already account for a striking share of national power use. Europe is trying to civilise the footprint—the Energy Efficiency Directive’s reporting rules, heat‑reuse pilots in Denmark, siting near renewables—but local patience is thinner than the press releases.

Here is the uncomfortable truth: compute is becoming industrial policy. If the US and Europe refuse to build the capacity their digital and scientific ambitions demand, others will—Gulf states courting hyperscalers with cheap land and power, Southeast Asian hubs expanding after Singapore’s partial thaw, India and China racing to sovereign compute at scale. A reflexive “no” will not slow AI; it will export it, along with the jobs, standards, and leverage that come with proximity to the machines. The better European question is not whether to build, but how to earn the consent to do so—tying permits to grid investment, water stewardship and heat recovery, clustering where renewables are abundant, and insisting on model accountability that makes pacing credible rather than performative. Business leaders who want Europe to lead in AI should show up in these local debates with facts, humility, and a willingness to share the upsides; otherwise, we will wake up to find that the future was built—just not here.