Weekly long-read: This week, I read an article about AI, discussing the question: 'Is there a bubble in AI?' When AI investments truly become excessive, who will ultimately bear the debt? According to estimates, between 2025 and 2032, cumulative investments in U.S. AI data centers, chips, electricity, and network infrastructure could reach a staggering $10.3 trillion, averaging 3.63% of the U.S. GDP annually. In terms of relative economic scale, this even surpasses historical waves of railroad, highway, electrification, and communication infrastructure construction in the U.S. So, can AI generate enough money to justify these investments? If not, who will bear the costs that have already been spent? When investments rely on cash flow, a bubble burst first hurts the shareholders. But when more and more investments start relying on debt, SPVs, private credit, and asset securitization, things change. Because when the next turning point in the AI market truly arrives, we might need to look beyond just the Nasdaq. Instead— we should watch whether cracks first appear in the credit market. If they do, those cracks might be the signal that this round of AI super capital expenditure cycle is truly entering its second half. Although there’s no sign of such risks yet, the risks are becoming increasingly harder to detect.