𝗧𝗵𝗲 𝗔𝗜 𝗦𝘂𝗯𝘀𝗶𝗱𝘆 𝗖𝗿𝗶𝘀𝗶𝘀
AI development moved from a dream to a daily tool in less than ten years. We went from text toys to models like Claude and GPT in a blink.
But the industry is hitting a wall.
The history of this shift follows four stages:
The Dream (2010-2016): People shared knowledge on Wikipedia and Stack Overflow. It was community driven. We moved away from this toward centralized systems.
The Disruption (2016-2021): Programmers grew tired of strict rules on Stack Overflow. When communities reject questions, people turn to AI. Google released the Transformer paper in 2017. Then COVID-19 forced everyone into isolation. These events created the perfect path for virtual assistants.
The Hook (2023-2025): ChatGPT made AI feel like magic. Tools like Copilot and Deepseek became essential. People believed AI was built to empower humanity.
The Endgame (2025-Present): The math is failing. VCs want profits now. AI companies optimized for the long term, but investors lose patience quickly.
The problem is simple: Frontier models run on subsidies.
A 20 dollar monthly subscription does not cover the cost of the massive compute power required to run these models. Companies are losing money on every user. To make a profit, they must charge based on every word generated. This makes AI expensive and hard to use.
AI tries to automate human reasoning and logic. This is the hardest task in history.
If companies stop the subsidies, users will leave. People might go back to Wikipedia or switch to open source models like Llama that run on local hardware.
Companies cannot subsidize these costs forever.
What happens when the money runs out?
Source: https://dev.to/prahladyeri/the-ai-conundrum-we-are-living-in-highly-subsidized-interesting-times-ldl
Optional learning community: https://t.me/GyaanSetuAi