After publicly shorting MU before, I’ve recently been shorting Hynix continuously.

I started with a principal of 2,000U and have gotten it to about 85% gains; I closed the position just now at 1340.
Let’s do a quick review.
In the 2008 real estate bubble, many people thought the housing prices would first collapse—then homeowners would default on their loans, and the crisis would explode.
But the more accurate order is:
Mortgage subprime defaults started increasing first, then the growth rate of home prices couldn’t keep up with leverage, and only then did the whole system blow up.
The core of it is:
From the beginning, this system wasn’t run on “normal loan repayments.”
It relied on housing prices continuing to rise.
When home prices go up, collateral values rise; loans can be rolled over, and risks can be hidden.
As long as asset prices keep accelerating, all problems can be delayed.
But once the growth rate slows down, even if prices haven’t crashed yet, cracks in the system have already started to show.
When I look at the AI industry chain now, I also get a similar feeling.
Semiconductors, memory, data centers, power, AI systems—this chain looks strong at every link right now.
Orders are real.
Demand is real.
AI may really change the world too.
But the problem is:
The market isn’t pricing today’s profits anymore—it’s pricing rapid growth for many years into the future.
Data centers that haven’t been fully built yet are already supporting a super-cycle for memory and chips.
AI revenues that haven’t fully materialized yet are already supporting compute contracts at the scale of several hundred billion dollars.
In the end, this entire chain needs a small number of top model companies and cloud providers to turn these compute demand into real revenue.
OpenAI, Anthropic, Google, Meta, xAI, Microsoft, Amazon…
These companies are not only the stars of the AI narrative; they’re also the ultimate payers for the upstream industry chain.
The issue is:
If in the future the revenue growth rate stays fast enough, the whole system can keep rolling.
If financing stays smooth, contracts keep getting signed, and capital keeps believing, then the upstream can keep rising.
But if one day the payers start saying:
“We want to slow down AI spending.”
Then the problem likely won’t be just that one company’s earnings are bad.
Instead, the entire AI pricing logic needs to be recalculated.
The real estate bubble lived on acceleration in housing prices.
An AI bubble may live on acceleration in compute demand.
A machine that survives on acceleration is most afraid not of an immediate reversal.
It fears a slowdown.
So I think the real signal that an AI bubble ends isn’t necessarily a certain chip company’s earnings report exploding.
It’s when the paying giants say they want to cut AI spending—and the stock market starts rising instead.
That would mean the market finally shifts from “growth faith” to “cash-flow discipline.”
Clearly, we may not be there yet.
Keep feeling the market’s fluctuations.