The Obscene Energy Demands of A.I

Net Energy Impact and Substitution

• Several comments argue you can’t judge AI’s energy use in isolation; you must model what activities it replaces (e.g., optimized supply chains, efficiency work) and compare marginal costs.
• Others find this argument unclear or speculative, noting a lack of solid quantitative models of such trade‑offs.

Comparisons with Crypto and “Obscene” Use

• Strong sentiment that Bitcoin’s proof‑of‑work is an “obscene” energy user, especially because difficulty adjusts to negate efficiency gains.
• Some distinguish Bitcoin from other crypto (e.g., proof‑of‑stake chains with low energy use) and caution not to conflate all crypto.
• Back‑of‑envelope estimates from the thread suggest ChatGPT uses ~3 orders of magnitude less electricity than Bitcoin while serving far more people.
• A minority sees the AI‑energy narrative as a recycled PR attack once used against crypto; others insist climate concern is genuine.

Efficiency, Models, and Hardware

• Many expect substantial long‑term efficiency gains via better hardware, pruning, specialized models, and open‑source optimizations.
• There’s tension between using huge LLMs for simple tasks versus smaller, task‑specific models (e.g., BERT‑style) that may be more efficient.
• Some argue on‑device open models might be more efficient; others note total energy is still consumed and distributed, not eliminated.
• One view: efficiencies will simply enable more AI usage, so total wattage may not fall (induced demand).

AI’s Practical Benefits vs “Fluff”

• Pro‑AI comments list concrete gains: demand prediction for power plants (megawatts saved), supply‑chain optimization, code and SQL generation, translation, dictation, quality control, medical imaging, forecasting, and rapid graphics for presentations/training.
• Skeptics see a lot of “fluff” (stock‑like images, bureaucratic automation, marginal convenience) and question whether current gains justify massive capex and emissions.
• There’s debate over whether these benefits are real productivity or hype subsidized by investors.

Energy Systems, Nuclear, and Renewables

• One camp claims there’s no meaningful “obscene” demand: civilization needs vastly more energy, the key is making it clean.
• Strong advocacy for nuclear (including thorium and advanced designs) as effectively zero‑impact and scalable, with critics countering with uranium mining, Chernobyl, Three Mile Island, waste longevity, and construction cost overruns.
• Others argue rising AI demand could accelerate deployment and cost declines of solar, wind, and batteries, thereby speeding decarbonization.
• EV adoption is mentioned as likely a far larger grid stressor than AI in the long run.

Markets, Policy, and Climate Framing

• Some insist markets, under good rule of law, equate profit with usefulness and governments should regulate pollution but not decide “good” vs “bad” energy uses.
• Opponents argue markets optimize profit, not social or environmental outcomes, and that relying on perfect government or perfect markets is unrealistic; individual responsibility and regulation remain necessary.
• There is disagreement over the urgency of climate change (“emergency for 50 years” vs recently salient) and whether appeals to “degrowth” are helpful.

Behavioral Effects and Induced Demand

• Multiple comments worry that AI’s convenience leads to over‑reliance (people asking chatbots instead of thinking), potential decline in analytical skills, and an eventual flood of AI‑generated garbage once energy costs drop.
• Others see this as a user‑education or cultural issue rather than an inherent problem with AI itself.