conv.

All stories
AIFading · day 6

AI output costs plummet 47% quarterly as price war intensifies

Analysis shows generative AI becoming cheaper faster than any transformative technology in history, outpacing chip shortage paradoxically.

newest coverage 7h agosummary written 10h ago7 pieces1.0k likes & upvotes61 comments BlueskyNewswiresHacker NewsMastodon

What to know

  • AI model output has become radically cheaper—down 47% per quarter over three years, with a 725-fold drop in some reasoning task costs in just 18 months.
  • A neural network researcher argues this spending growth may be driven by structural economics: unlike traditional software, AI systems have no saturation point and can always theoretically benefit from more compute.
  • The price collapse creates a paradox: AI chips and electricity remain scarce and expensive inputs, yet the output users buy is deflating faster than almost any technology in history.

The dispute Whether rapid cost-per-performance improvements validate the economic case for AI investment or mask an underlying incentive structure (unlimited capacity absorption of compute resources) that ensures spending continues regardless of practical gains. · positions read across 2 posts and comments

many voices

AI price collapse reflects real technological progress and efficiency gains in model training.

  • “It is like the sticker price on a new car falling from $50,000 to $69.”

    Epoch analysis · Epoch AI ↗
some voices

Massive AI spending is driven by structural economic incentives to consume compute, not demonstrable practical progress.

  • “One which I have written about recently is the idea that it's all a cover story for stockpiling chips in case China invades Taiwan in 2027.”

    rossdavidh · Hacker News ↗

Ross David Harthorn Neural networks researcherEpoch Research organizationOpenAI AI model developer

AI output costs plummet 47% quarterly as price war intensifies
epoch.ai

How it unfolded 2 developments, newest first · click a bar or a number to jump articlesposts

Peak 1 piece in two hours at Sep 17, 6 PM; 7 pieces over 6 days (2 articles · 5 posts) Sep 17, 6 PM — 1 piece · 1 post — Bluesky 1Sep 17, 8 PM — quietSep 17, 10 PM — quietSep 18, 12 AM — quietSep 18, 2 AM — quietSep 18, 4 AM — quietSep 18, 6 AM — quietSep 18, 8 AM — quietSep 18, 10 AM — 1 piece · 1 article — Newswires 1Sep 18, 12 PM — quietSep 18, 2 PM — quietSep 18, 4 PM — quietSep 18, 6 PM — quietSep 18, 8 PM — quietSep 18, 10 PM — quietSep 19, 12 AM — quietSep 19, 2 AM — quietSep 19, 4 AM — quietSep 19, 6 AM — quietSep 19, 8 AM — quietSep 19, 10 AM — quietSep 19, 12 PM — quietSep 19, 2 PM — quietSep 19, 4 PM — quietSep 19, 6 PM — quietSep 19, 8 PM — quietSep 19, 10 PM — quietSep 20, 12 AM — quietSep 20, 2 AM — quietSep 20, 4 AM — quietSep 20, 6 AM — quietSep 20, 8 AM — quietSep 20, 10 AM — quietSep 20, 12 PM — quietSep 20, 2 PM — quietSep 20, 4 PM — quietSep 20, 6 PM — quietSep 20, 8 PM — quietSep 20, 10 PM — quietSep 21, 12 AM — quietSep 21, 2 AM — quietSep 21, 4 AM — quietSep 21, 6 AM — quietSep 21, 8 AM — quietSep 21, 10 AM — quietSep 21, 12 PM — quietSep 21, 2 PM — quietSep 21, 4 PM — 1 piece · 1 post — Hacker News 1Sep 21, 6 PM — quietSep 21, 8 PM — quietSep 21, 10 PM — quietSep 22, 12 AM — quietSep 22, 2 AM — quietSep 22, 4 AM — quietSep 22, 6 AM — quietSep 22, 8 AM — quietSep 22, 10 AM — 1 piece · 1 article — Newswires 1Sep 22, 12 PM — quietSep 22, 2 PM — quietSep 22, 4 PM — 1 piece · 1 post — Hacker News 1Sep 22, 6 PM — quietSep 22, 8 PM — quietSep 22, 10 PM — quietYesterday, 12 AM — quietYesterday, 2 AM — quietYesterday, 4 AM — quietYesterday, 6 AM — quietYesterday, 8 AM — quietYesterday, 10 AM — 1 piece · 1 post — Hacker News 1Yesterday, 12 PM — quietYesterday, 2 PM — quietYesterday, 4 PM — 1 piece · 1 post — Mastodon 1Yesterday, 6 PM — quietYesterday, 8 PM — quietYesterday, 10 PM — quiet 12
Sep 18Sep 19Sep 20Sep 21Sep 22yesterdaynow · 12:58 AM ET
  1. 2

    Epoch reports AI output costs fell 47% quarterly over three years

    Analysis of five AI performance benchmarks spanning mathematics, hard sciences, and game-based puzzles shows a 47% per-quarter cost decline for a given level of performance. The study highlights a 725-fold price drop in 18 months: OpenAI's o3 model (January 2025) achieved a 75% score on GPQA Diamond for 30 cents per question; GPT-5.6 Luna achieved the same score for $0.0004 per question. The authors note that initial performance breakthroughs command premium pricing before competition drives costs down.

    “No other general-purpose technology in history appears to have gotten so cheap so fast.”
    — Epoch analysis
    • remixtures@tldr.nettime.org

      "AI has gotten cheaper more quickly than any other transformative technology in history. The cost of achieving a given level of AI performance has fallen about 47% per quarter since 2023, or 13× per year. That price drop is four times faster than DNA sequencing, six times faster than compute, 18 times faster than lithium batteries, and (in the…

      remixtures@tldr.nettime.orgMastodon7h agoview on Mastodon ↗
  2. 1

    Researcher questions economics of AI spending

    Ross David Harthorn, with a background in neural networks since the mid-1990s, argues that unlike conventional software, neural networks have no point of saturation—they can always use more hardware, training data, and epochs. He suggests this structural incentive to buy ever more compute (combined with the ambition of reaching AGI rather than solving specific tasks) may explain why so much money flows into AI regardless of practical performance gains.

    “Neural networks, are not like that. They never say 'no thanks, I'm full'. You can always add more neurons, more hidden layers, more epochs of training, or more data to train on.”
    — Ross David Harthorn, Neural networks researcher · source

What people are saying 0 voices from 0 sites · best of 2 · verbatim