Take a step back and look at the current situation of AI, and add a bit of self-comfort.

My biggest concern remains whether land, electricity, and data center shells can be built fast enough to absorb the computing power supply. This is a kind of "mild bullishness within a bearish view": in the short term, some companies' numbers may fall short of expectations, but the cycle will be prolonged—same old story.Currently, I guess it's sufficient, but I say that with humility. The amount the rest of the world can absorb is actually more than people think (Latin America has become a new hotspot), but construction is very complex and time waits for no one.However, last night's ANT news and today's TSMC news have made me a bit more optimistic.1) TSMC is extremely conservative; they don't jump on bandwagons easily. They sent people on-site to verify the data centers weren't fake news, and are even raising capital expenditure with confident messaging. I take this much more seriously than those so-called industry experts just staring at satellite images. My apologies if this offends anyone.2) The current economics of frontier labs make even more sense than I previously thought. Earnings reports and statements on X indicate that, even after deducting training costs, profit margins are decent, and (for now) free cash flow is positive. That’s pretty impressive.This is all performance since the start of the year, which includes two smaller, lower-margin quarters, plus the outrageous computing prices paid to Elon. (Too bad Elon isn't as generous with himself as he hopes SK Hynix will be...)As non-Elon capacity comes online, computing costs should come down. The headwinds from compliance and distillation will soon pass too.3) The implied ARR figure is lower than many people think (though in line with what analysts I regularly talk with estimate). Mixing up gross/net, poor reporting, fake news... overall, it’s quite annoying.But imagine if ARR reaches $85 billion, how differently everyone will sound. Judging by the computing power coming online and the hype around the 5.5 series in public data, it may already be there, or will be in a few weeks.So will the target figure be hit just 4–6 weeks late? If that spooks you, then you should have been even more afraid at almost every point in the past two years—especially this June!By the way, back then, no one (myself included) expected they'd withstand the combined blows of safety, Microsoft/Meta, and distillation all at once.All in all, I think market sentiment will shift very soon.