November 15, 2025 12:04 PM (IST)
Date first published: November 15, 2025 11:13 AM IST
Written by Sai Rahul Poluri
A German court ruled this week that OpenAI, the creator of ChatGPT, must pay undisclosed damages for using copyrighted music without the permission of the affiliated organization. In September, Anthropic, the creator of Claude, had to pay $1.5 billion to authors whose works were pirated to train AI models. Anthropic’s settlement is in response to a class action lawsuit alleging that more than 7 million books were pirated by Anthropic, and the OpenAI lawsuit was brought by a group representing 100,000 lyricists, music producers, and composers. It is worth noting that Anthropic could theoretically be liable for up to $1 trillion in damages, 1,000 times the amount agreed to in the settlement. It’s also worth comparing this figure to the almost $10 billion that Anthropic is expected to generate in revenue in 2025.
How much will GenerativeAI models cost if authors have to pay worldwide royalties for the copyrighted works used to train their respective models? The cheapest non-free plans for OpenAI ChatGPT and Anthropic Claude cost Rs 399 per month and $17 per month, respectively. If Anthropic were forced to pay $100 billion instead of $1.5 billion for pirated books, would it be forced to raise the price of Claude to $100 per month to recoup its losses? Could OpenAI offer a free tier of ChatGPT to keep users hooked? Would it continue to subscribe to Claude or other large-scale language models (LLMs) if the costs were orders of magnitude higher?
GenerativeAI’s current business model is hyper-exploitative and dependent on taxpayer subsidies. The land used to build data centers is usually provided to companies on a subsidized basis. Data centers use large amounts of fresh water to cool the power-hungry chips that run LLM, and even more is consumed indirectly, such as through power generation. Should the cost of fresh water to humans be the same as it costs to machines? Data centers place a significant strain on the power grid, and increased demand typically causes electricity costs near data centers to rise. Governments can choose to provide electricity at subsidized rates, but should the world’s richest corporations do the same? Should data centers be required to use chips and other electronic components made from rare earth elements sourced from conflict zones?
As beautifully depicted in the movie Humans in the Loop, LLM relies on cheap labor to label, annotate, and perform other tasks necessary to prepare data for model training. Additional inexpensive labor is required to verify and adjust the quality of LLM output. And last but not least, most LLMs train in copyrighted books, images, music, videos, code, and other creative works without paying the required royalties to the creators. Companies are subsidized directly or indirectly at every step of the AI supply chain. How much would an LLM cost if we took away all of these subsidies? And we can’t even imagine the opportunity cost of focusing on LLM and GenerativeAI at the expense of everything else, especially when an MIT study reported that 95 percent of enterprise AI pilot projects did not produce meaningful results.
Silicon Valley’s attitude of asking for forgiveness instead of permission goes beyond copyright law. Do we want to tolerate water shortages, unstable power grids, and exploitative labor markets?
(The author is FOSS United CEO)
