Amazon buys stake in nuclear energy developer in push to power data centres

Original Article ↗ Hacker News Discussion ↗

Rising power demand from AI data centers is pushing Amazon and other tech giants toward small modular nuclear reactors as a source of “firm” low‑carbon electricity built near their campuses. Commenters clash over whether nuclear is economically and operationally superior to rapidly scaling wind, solar and batteries, citing capacity factors, lifetime costs, subsidies, waste handling and examples from France, Germany, China and the US. Many welcome a diversified grid with both renewables and nuclear, but skeptics warn that SMRs remain largely unproven, may arrive too slowly and too expensively, and risk diverting money from solutions that are already being deployed at scale.

Scope of the announcement

  • AWS talked about plans for ~5 GW of small modular reactors (SMRs), joining other big tech nuclear moves (Google, Microsoft, Oracle, Sam Altman, Gates, etc.).
  • Some see this as a meaningful financial commitment with real downside risk; others note current investments are small compared to the eventual multi‑tens‑of‑billions needed.

Nuclear vs. solar/wind + storage

  • Pro‑nuclear side:
    • Argues AI/data centers need gigawatt‑scale, 24/7, low‑carbon power that wind/solar can’t reliably provide due to intermittency.
    • Claims batteries can’t economically cover multi‑day or “arbitrary length” weather shortfalls; storage requirements would explode as renewable penetration rises.
    • Notes existing nuclear fleets run at ~93–96% capacity factor and can be planned for refueling/maintenance, unlike weather.
  • Pro‑renewables side:
    • Cites rapidly falling costs and huge global build‑out of solar, wind, and batteries; nuclear additions are tiny by comparison.
    • Points to modeling showing fully renewable systems (with flexibility and storage) cheaper overall than nuclear‑heavy systems, with nuclear needing ~80–85% cost reduction to compete.
    • Argues demand can be shifted and grids can handle intermittency with forecasting, interconnection, demand response, and diverse sites.

Reliability and grid behavior

  • One camp emphasizes nuclear’s very high fleet‑wide capacity factors and uncorrelated outages, calling it the most reliable large‑scale generation.
  • Critics counter with:
    • Long unplanned outages at individual plants.
    • Examples where many reactors are simultaneously offline.
    • Issues like heat‑wave cooling constraints and political shutdowns.
  • Consensus: nuclear is not 100% available; renewables are highly variable. Disagreement is about which mix gives the most dependable system at lowest cost.

Economics and subsidies

  • Nuclear advocates: once built, plants “print money”; fuel is cheap; long lifetimes amortize capex. They blame nuclear troubles on policy, stop‑start build programs, and distorted markets (priority, subsidies, and negative prices for renewables).
  • Skeptics: highlight cost overruns (Vogtle, Hinkley, Flamanville), bailouts and renationalizations, special tariffs, and underfunded waste management as evidence nuclear isn’t truly profitable without heavy state support.
  • Both sides accuse the other of ignoring system‑level costs (transmission, backup, curtailment, waste, decommissioning).

Climate and technology strategy

  • One faction: “If you’re not serious about nuclear, you’re not serious about climate,” citing Germany’s nuclear exit vs. France’s low‑carbon mix.
  • Opponents say Germany is decarbonizing via renewables and that market trends overwhelmingly favor solar/wind+batteries; nuclear’s slow deployment (10–15+ years) makes it marginal for urgent climate timelines.
  • Alternatives discussed: hydro, pumped storage, geothermal, tidal, advanced reactors (breeders, molten salt, thorium), but their scalability and maturity are debated.

Data centers and power

  • Strong synergy argued between nuclear and data centers: steady, high, year‑round load; shared expertise in high‑reliability complex systems.
  • Others note DCs still need grid interconnection and that, today, GPUs are far more expensive than electricity, so even costly grid power plus renewables can make economic sense.
  • Some foresee a split: nuclear‑backed “elite” power for AI and industry vs. cheaper, variable renewables for general consumers.