You can always look at AI, or rather, the question of AGI, from two completely different perspectives.
From a technical perspective, the unprecedented prospect you face is that AGI is on the horizon. Every advance you see from reinforcement learning supports the idea that AI can start from scratch and quickly surpass human intelligence, not just in one or two domains—not only in chess, Go, protein structure prediction, or conversational bots and multimodal large language models—but rather, one model that can handle a sufficient number of domains.
The constraint you face, perhaps, is that data and computing power are far from enough. For a long time in the past, whenever results were poor during algorithm research, after questioning yourself "Is the method wrong?" for only a few months, you would always get surprisingly amazing results after adding more data and more computing power—the bitter lesson.
You also need to face the question: is human "regulation" failing to keep up with the pace of AI development?
From a financial perspective, a question becomes increasingly direct: what is the return on continuously increasing investment, adding computing power, and driving the Capex of the entire AI infrastructure?
If the issue is merely a financial challenge, only considering whether it's a financial bubble or a question of over-investment, it wouldn't be a particularly big problem. Although the burst of a bubble would lead to a huge drop in market capitalization, the players at the table in this AI game are very limited. As long as you believe AGI is certain to arrive, the return rate will not be low in the long run.
But what if surface intelligence simply cannot lead to AGI in a short time? We will still face this dilemma for a long period: AI demos and even tools for personal service work very well, but they just can't be deployed to production environments at scale.
What if the data gap is ten or more orders of magnitude?
Will an AI Winter arrive?