According to Amazon's data, data centers now basically operate on 5-6 year long-term contracts + 3 years to break even. To be honest, this return... why does it sound worse than mining $BTC? Mining carries the risk of $BTC price fluctuations, while data centers bear the risk of demand fluctuations. They say it's a long-term contract, but when major changes happen, those contracts can't hold up. By the end of 2025, ever since Anthropic discovered the coding use case, AI hasn't really uncovered any good new use cases for over half a year. Long-term prospects are likely bright, but short-term demand growth is indeed highly volatile. I heard A-company's ARR growth has already slowed down. Plus, the coding use case has already been price-crushed by Bharat's models like DS and Kimi. Why can't we find new high-value use cases? Because the models are dumb and their evolution speed isn't fast enough. Here's a random example: why do doctors still exist for consultations? Because even though large models have massive medical knowledge, in practice, their diagnostic abilities are still not as good as doctors. Personally, I'm actually disappointed with the intelligence progress of large models over the past six months—it's clearly below my expectations. The future is definitely bright.