November 15, 2025 11:13 AM IST First published on: Nov 15, 2025 at 11:13 AM ISTA German court ruled this week that OpenAI, the creators of ChatGPT, must pay undisclosed damages for using copyrighted music without permission from relevant organisations. In September, Anthropic, the creators of Claude, had to pay $1.5 billion to authors whose work was pirated to train its AI models. The Anthropic settlement was in response to a class-action lawsuit that alleged that over 7 million books were pirated by Anthropic and the OpenAI lawsuit was filed by an organisation representing 1,00,000 lyricists, music producers, and composers. It’s worth noting that Anthropic was theoretically liable for up to $1 trillion in damages, a 1,000 times more than what they agreed to settle for. It’s also worth comparing that number with the almost $10 billion that Anthropic is expected to generate as revenue in 2025.What would the GenerativeAI models cost if the creators had to pay worldwide royalties for the copyrighted works that are used to train their respective models? The cheapest non-free plans of OpenAI ChatGPT and Anthropic Claude cost Rs 399 per month and $17 per month respectively. If Anthropic was forced to pay $100 billion instead of $1.5 billion for the books they pirated, would they be forced to raise Claude prices to $100 per month to recoup their losses? Would OpenAI be able to provide a free tier of ChatGPT to get users hooked? Would you continue your subscription to Claude, or any other Large Language Model (LLM), if the costs were an order of magnitude higher?AdvertisementCurrent business models for GenerativeAI are hyper-extracative and rely on taxpayer subsidies. The land used to construct data centres is usually provided to the company at a subsidised rate. Data centres use enormous amounts of freshwater to cool the power-hungry chips that run the LLMs, and even more freshwater is consumed indirectly, for example, via electricity generation. Should freshwater cost the same to a human as it does to a machine? Data centres put incredible strain on the electricity grid, and electricity rates near a data centre usually go up because of the increased demand. Governments can choose to provide electricity at a subsidised rate, but should this be the case for the richest companies in the world? Should the data centres be required to use chips and other electronic components that are made with rare earth elements sourced from conflict-free zones?As beautifully portrayed in the movie Humans in the Loop, LLMs rely on cheap labour for data labelling, annotation, and other tasks necessary to prepare data for model training. Additional cheap labour is required to verify and tune the quality of the LLM output. And last but not the least, most LLMs are trained on copyrighted books, images, music, videos, code, and other creative works without paying the necessary royalties to the creators. The companies are subsidised, directly or indirectly, at every step of the AI supply chain. How much would a LLM cost if all of these subsidies were removed? To top it all off, we can’t even begin to imagine the opportunity cost of focusing on LLMs and GenerativeAI at the cost of everything else, especially when an MIT study reports that 95 per cent of Enterprise AI pilot projects delivered no meaningful results.The Silicon Valley attitude of asking for forgiveness instead of getting permission goes beyond copyright law. Do we want to forgive water scarcity, unstable electricity grids, and an exploitative labor market?(The writer is CEO, FOSS United)