Datacentres drive up big tech's carbon emissions to a third of those of France

Original Article ↗ Hacker News Discussion ↗

Scale of data center emissions and electricity use

  • Datacenters’ emissions are framed as “huge”, with some noting they now reach roughly a third of France’s emissions.
  • Context: France already has relatively low‑carbon electricity due to nuclear, so the comparison may undersell the impact.
  • Irish stats cited: datacenters now use 23% of Ireland’s metered electricity, up from 5% in 2015, which many see as extreme, especially given high local power prices.

Grid constraints, interconnection, and who pays

  • Several comments highlight long interconnection queues and slow grid studies as a bottleneck for adding new generation (including solar).
  • There is disagreement: some say new solar connects rapidly; others say that hides a massive backlog.
  • Some claim datacenter developers avoid paying for proper grid upgrades and externalize costs; when required to pay upfront fees, many projects reportedly vanished.
  • Datacenters increasingly install onsite fossil generation (diesel, gas turbines) due to grid and permitting constraints, raising “Scope 1” emissions.

Policy ideas: nuclear, renewables, and carbon pricing

  • Strong support from some for massive nuclear build‑out plus high CO₂ prices pegged to real carbon removal costs.
  • Others stress that political trends (regulatory dismantling, right‑wing backlash) make robust carbon pricing and regulation unlikely.
  • Suggestions:
    • Require datacenters to over‑provision solar and batteries, acting as grid anchors.
    • Mandate carbon‑free or at least non‑CO₂‑positive power for new facilities.
    • In the EU, some note datacenters are within the emissions trading system, sharing a shrinking cap.

AI demand, “needs,” and degrowth

  • Debate over whether society “needs” current and future levels of AI compute, or if it’s mostly frivolous or harmful usage.
  • Some argue compute demand is effectively unbounded and we should maximize clean electricity (even invoking Dyson‑sphere‑scale thinking).
  • Others worry datacenters crowd out existing consumers and worsen climate impacts; some advocate delaying or restricting new AI datacenters until they are cleanly powered.
  • There is tension between calls for “degrowth” in energy use and arguments that progress, like hot/cold water or AI, naturally expands consumption.

Economics, subsidies, and bubble risk

  • Disagreement on whether AI inference is subsidized:
    • One side says large providers are unprofitable and subsidized by venture capital.
    • Others point to low‑cost models and independent APIs as evidence that at least lower‑tier inference can be profitable and cheap.
  • Concerns raised about financial structures (e.g., index inclusions of AI‑related firms) channeling retirement savings into a potential AI bubble whose collapse could have broad impacts.

Skepticism about climate policy tools

  • Many are skeptical of “carbon neutrality” claims and carbon offsets, seeing them as largely cosmetic.
  • Some mock small‑scale consumer regulations (e.g., straws, bottle caps, plastic fees) as inadequate or symbolic compared to unchecked growth in datacenter emissions.