Self-Hosting & The Future of AI
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
Claude model activity is surfacing — worth tracking for capability changes, ecosystem impact, and availability.
The high cost of large language models (LLMs) is a significant concern, with heavy users potentially spending thousands annually, despite some free plans. The industry heavily subsidizes current AI tool usage, as evidenced by substantial losses from companies like OpenAI and Anthropic in 2024, and Perplexity's high spending on AWS and other AI providers. This financial strain suggests a "desperate race for survival" within the AI sector, pointing towards self-hosting as a potential future solution, though not yet fully viable.
LLMs are expensive. It may seem at first that paying $20 a month is a relatively cheap price for the value proposition, but this compounds to $240 per year. Heavy users who require more sophisticated or more integrative tools may pay as much as $200 a month, compounding to $2,400 a year. Yet, most providers offer free use plans, which, limited as they may be, preserve some of the equality required by this tool.
Yet, our use of these tools is still very heavily subsidized by the industry, OpenAI lost $5B in 2024, Anthropic $5.3B, Perplexity spends 164% of their revenue on AWS, Anthropic, and OpenAI. This is a desperate race for survival. But there may be a better way, and self-hosting may be the future of AI, just not yet.
Read more on my newest substack post:
https://pedrorodriguesribeirophd.substack.com/p/self-hosting-and-the-future-of-ai?r=9040kf&utm_campaign=post&utm_medium=web