We May Witness the ChatGPT Moment for Agent in 2025

Disclaimer: The content provided in this newsletter is for informational purposes only and does not constitute investment advice. We are not registered investment advisors, and nothing in this newsletter should be construed as a recommendation to buy or sell any securities. Always do your own research and consult with a licensed financial professional before making any investment decisions.

I have previously written a few articles about MSFT's delayed IDCs, but I must clarify that the power consumption of these IDCs is in the backlog, not in the already-built IDCs. It is more about lowering expectations for future CAPEX. However, now with the crazy growth of Claude 3.7’s ARR, I am starting to feel less pessimistic about CAPEX and even feel that CAPEX is a lagging indicator.

I will explain in more detail what impact Anthropic will have on AWS tomorrow.

Let's start by explaining Agent first.

Pretraining has not ended yet

The limitations of Scaling Laws suggest that the prevailing paradigm, which has leaned heavily on model size, data volume, and raw compute, may be approaching its limit. The era of simply scaling Transformer architectures on readily available internet data to boost model capabilities is largely behind us, as that well of public data is nearly dry. However, pre-training itself is far from over. Leading research labs are now truly beginning to explore the right approaches. The key is to discover new, more efficient scaling laws, tailored to specific application demands—integrating diverse needs related to compute, data, and agent functionalities.