Why AI Is Getting More Expensive, Not Cheaper

In January 2025, one PhD-level science question cost about 30 cents to answer with a top AI model. Less than 18 months later, a newer model answered it just as well for four hundredths of a cent. So why are AI prices, office software and even electricity bills going up? The answer is a 160-year-old idea about coal: when something gets much cheaper to use, we don't use less of it. We follow the price of an answer all the way down, from the token to the chips, the data centers and the power grid. This video is for education only and is not financial advice. Chapters: 0:00 A $0.30 question 1:18 The collapse 2:19 1865 4:21 The flood 5:17 Thinking costs 6:54 The price comes back 8:38 The machine 10:08 Who pays 12:19 Cheaper and more expensive 13:21 The real reason Sources: Epoch AI, The plunging price of thought (Sep 2026): https://epoch.ai/publications/the-plu... Epoch AI, LLM inference price trends: https://epoch.ai/data-insights/llm-in... Epoch AI, output length of reasoning models: https://epoch.ai/data-insights/output... a16z and OpenRouter, State of AI (100 trillion token study): https://arxiv.org/html/2601.10088v1 OpenAI API pricing: https://developers.openai.com/api/doc... Google Gemini API pricing: https://ai.google.dev/gemini-api/docs... Google I/O 2026 keynote (tokens per month): https://blog.google/innovation-and-ai... Alphabet Q2 2026 earnings remarks: https://blog.google/company-news/insi... ChatGPT weekly users (TechCrunch): https://techcrunch.com/2026/02/27/cha... GitHub Copilot usage-based billing: https://github.blog/news-insights/com... Cursor pricing update (2025): https://cursor.com/blog/june-2025-pri... Microsoft 365 pricing update (July 2026): https://www.microsoft.com/en-us/licen... OpenAI finances, as reported by the Financial Times: https://www.thestar.com.my/tech/tech-... Anthropic Series H announcement: https://www.anthropic.com/news/series-h Microsoft, Alphabet, Amazon and Meta 10-K and 10-Q filings (capital expenditures), via SEC EDGAR. NVIDIA Q2 fiscal 2027 results: https://nvidianews.nvidia.com/news/nv... Lawrence Berkeley National Laboratory, United States Data Center Energy Usage Report: https://eta-publications.lbl.gov/publ... U.S. EIA, Short-Term Energy Outlook (Oct 2026): https://www.eia.gov/outlooks/steo/rep... PJM capacity auction results: https://insidelines.pjm.com/pjm-aucti... and https://insidelines.pjm.com/pjm-aucti... DC Office of the People's Counsel (May 2025): https://opc-dc.gov/wp-content/uploads... TrendForce DRAM contract prices: https://www.trendforce.com/presscente... Raspberry Pi price update (Oct 2026): https://www.raspberrypi.com/news/pric... Xbox console price update (Jun 2026): https://news.xbox.com/en-us/2026/06/2... W. S. Jevons, The Coal Question (1865), chapter VII. UK coal output, NBER Macrohistory via FRED: https://fred.stlouisfed.org/data/A012... Fouquet and Pearson, The Long Run Demand for Lighting (2012): https://eeep.iaee.org/?p=1449 Nvidia market value loss on Jan 27, 2025: Reuters, LSEG data. Satya Nadella on Jevons paradox (Fortune, Jan 27, 2025): https://fortune.com/2025/01/27/micros... Photos and drawings, all via Wikimedia Commons: Sam Altman speaking at TED by Steve Jurvetson (CC BY 2.0). Raspberry Pi 5 by SimonWaldherr (CC BY 4.0). Xbox series X by Ian Hughes (CC BY 2.0). GPU dies by Fritzchens Fritz (CC0). Data centers in Ashburn by Theodore Christopher (CC0). Electricity pylons at sunset in Quebec City by Wilfredor (CC0). William Stanley Jevons, 1858, and Popular Science Monthly, 1877 (public domain). Boulton & Watt steam engine, R. H. Thurston, 1878 (public domain). The Collier, George Walker and R. & D. Havell, 1814 (public domain). Thumbnails: Asus Strix RTX 4090 by Benlisquare (CC BY-SA 4.0) and Hydro quebec meter by Kristoferb (CC BY-SA 3.0). Brand sound effects: Magnific (commercial license). #ai #economics #technology