The cost of using AI is still 10-20 times cheaper than reality. When will AI companies stop subsidizing token prices?

The essence of AI is technology that reduces prediction costs (prediction). When the input prediction gets cheaper over time, businesses will gradually turn every decision into a probability problem.
Right now, only 0.28% of users worldwide are willing to pay for using AI.
Most paid AI users do so through (Plus) subscriptions, so the cost is already bundled into the (package not charged per token), and people hardly realize how expensive AI really is once you account for it.
On average, a basic Claude user, who doesn’t burn tokens on embedding, RAG, streaming data/crawl..., will burn about ~1-1.5M tokens until hitting the limit (, which is around 5 hours of tasks).
The weekly usage cap averages about 9-10M tokens per month for Claude. The cost for 1M tokens of Claude Sonnet is $3 (input) and $15 (output); if you use CLaude Opus, it goes up to $5-15 (input) and $25-75 (output).
-> This means if you use the API instead of the plus package, you’ll burn at least $200 for comparable performance(—without even counting the fact that you’re using Opus).
AI company subscriptions don’t sell tokens; they sell an experience of “no need to think about tokens” until you use too much. Soon, AI companies will increase token prices to make a profit, and these AI costs will gradually become fixed costs for everyone.
The title of this article is a bit clickbaity, but it’s clear you’re at the best time to tap into AI’s performance—you are a super early bird user.