The damage started with Alphabet, spread through Intel and Amazon, and by the time Asia opened the shock had made its way straight into Samsung Electronics and SK Hynix. The KOSPI opened sharply lower, SK Hynix was hit hard, Samsung followed, and the semiconductor complex suddenly found itself trading less like the beneficiary of an unstoppable AI infrastructure boom and more like collateral damage from the people paying the bill.
Takeaways by Dark Side of the Boom™
• Alphabet, Intel and Amazon have turned the AI return-on-capital debate from a Wall Street argument into an Asia opening problem, with Korea’s semiconductor complex taking the first hit.
• SK Hynix and Samsung are being sold not because HBM demand vanished overnight, but because investors are questioning how long hyperscalers can keep feeding the capex furnace at the current pace.
• Intel is the perfect contradiction inside the cycle: AI is boosting data-centre demand while memory scarcity and heavy investment are simultaneously choking the legacy business and free cash flow.
• Korea now becomes the overnight stress test for the entire AI chain. The real question is whether this is sentiment contagion or the first sign that investors are marking down the terminal value of the buildout itself.
The AI Boom Eats Its Own Tail in Asia
Wall Street lit the fuse, and Seoul is where we are finding out how far the blast travels.
That is what makes the move interesting.
Nothing fundamental in the HBM order book changed overnight. SK Hynix did not wake up to discover that customers had cancelled their long-term commitments, nor did Samsung suddenly lose its place in the memory cycle. What changed was confidence in the people writing the cheques further downstream.
Alphabet’s collapse put a price on the idea that AI capex can outrun monetisation. Intel then added a different kind of headache, showing that strong demand does not help much if the company cannot turn it into clean economics. Amazon followed with another reminder that cloud growth can still look healthy while free cash flow starts disappearing into the infrastructure furnace.
The boom is beginning to eat its own tail.
AI spending creates demand. Demand creates scarcity. Scarcity creates pricing power. Pricing power pushes the cost of the buildout even higher. Then Wall Street turns around and asks the hyperscalers why the buildout has become so expensive.
That loop has been hiding in plain sight for months.
SK Hynix looked almost untouchable only a day ago, sitting at the sweet spot of the cycle with memory sold years ahead, extraordinary margins and customers effectively queuing for supply. Now the market is asking a more uncomfortable question: what happens to the toll collector if the traffic stops accelerating?
That does not mean the traffic is disappearing. Far from it. The underlying demand for advanced memory remains powerful, and the AI infrastructure buildout is still absorbing capacity at a rate that would have looked absurd a few years ago. But markets do not trade only on whether demand exists. They trade on the rate of change, and the rate of change in confidence around AI spending has clearly turned.
Intel is the best example of how messy that becomes.
The company finally has the data-centre demand it wanted, but the same AI cycle is driving memory prices higher, making PCs more expensive, hurting the legacy business and forcing another round of heavy capital spending just as investors thought the turnaround was finally becoming self-financing. Intel is standing in the middle of the gold rush with customers at the door and still managing to trip over the boxes on the way out of the warehouse.
It is an execution problem wrapped inside a capital-intensity problem.
And once the market starts looking at the AI trade through that lens, every company gets measured differently.
SK Hynix can still make extraordinary money because scarcity sits on its side of the table. Samsung can still benefit from memory pricing and server demand. Microsoft can still show that cloud growth is strong enough to justify the spend. But Alphabet has shown that a dominant franchise can be punished brutally when monetisation starts lagging the capex curve, and Intel has shown that exposure to the boom is not enough if the rest of the business keeps leaking.
That is why Korea matters so much this morning.
The KOSPI has become an overnight seismograph for the global AI trade. When the US megacaps wobble, the first tremors run straight through memory, semiconductors and Korean large caps because those are among the cleanest listed expressions of the infrastructure buildout. If the confidence shock is temporary, the opening flush should eventually attract buyers back into the names where earnings remain intact. If the market is beginning to question the terminal pace of AI capex itself, then Korea has a bigger problem because the entire valuation structure rests on the assumption that spending keeps climbing.
The first move lower, therefore, tells us less than the response to it.
If foreign selling keeps accelerating and the market continues treating Samsung, SK Hynix and every other AI-linked name as one giant risk bucket, then this becomes positioning first and fundamentals second. That is usually when otherwise good companies get sold simply because somebody needs liquidity.
But if the market starts distinguishing between the companies whose economics are actually deteriorating and those merely caught in the blast radius, then the session becomes much more interesting.
That is the point where bargain hunters start showing up.
SK Hynix entered this session with extraordinary earnings momentum, long-term supply visibility and pricing power that has not magically disappeared because Alphabet had a bad quarter. Samsung is hardly trading a broken memory cycle either. The selloff is real, but the fundamental argument underneath it remains much more nuanced than the index move suggests.
This is why I would resist the temptation to call the entire AI trade broken.
What is broken is the old assumption that every part of the chain deserves the same multiple simply because the money is flowing through it.
The market is finally separating the suppliers from the spenders, the cash generators from the cash burners, and the businesses with genuine scarcity rents from those still trying to prove that capital intensity eventually turns into something shareholders can bank.
That is a healthier market in the long run.
It is a much more violent one in the short run.
For Asia, that means the easy thematic trade has given way to stock selection, flow watching and a much sharper focus on who actually captures the economics of the AI buildout.
Now Seoul gets to decide whether this is contagion or opportunity.
In the meantime, last week's Tech Wreck muscle memory is kicking back in: buy the yen, sell the gold. It looks counterintuitive only if you treat every risk-off episode as the same animal. When the stress is equity-led and liquidity starts tightening, JPY still behaves like the cleaner funding unwind, while gold can become the nearest ATM as traders raise cash and cover losses elsewhere. When tech starts falling through the floor, traders do not reach for a philosophy book. They reach for the hedges they remember.