I was reborn. This time, I woke up before the U.S. market opened on June 5. People in the chat were still debating whether AI semiconductors were going to keep flying, but I already knew what would happen in the next few hours.
So I did not hesitate. Before the open, I sold all my AI semiconductor stocks, every last one: Nvidia, Broadcom, Micron, Marvell. Then I flipped and shorted semiconductors.
Because I knew that on this day, U.S. stocks would teach everyone chasing AI semiconductors the most expensive risk lesson of all.
The entire U.S.-listed chip sector lost about $1.3 trillion in market value in a single day. How huge is that? Put simply, in just one day the market ripped an economy-scale chunk of value out of AI semiconductors.
Even more dramatic, the Philadelphia Semiconductor Index fell more than 10% in one session. Marvell dropped nearly 17%, Micron fell more than 13%, and AMD, Intel, Broadcom, and Nvidia were all dragged down too.
One second they were the stars of the AI bull market, with everyone believing compute never sleeps, chips never stop rising, and AI infrastructure is the printing press of the new era. The next second, an epic selloff swept Wall Street and U.S. stocks saw a terrifying "Black Friday."
In fact, the core reason Micron and Marvell plunged was not that demand vanished overnight, but that Wall Street suddenly realized the most crowded part of the AI chain is often the fastest place to get valuation crushed.
This round of selling was not a fundamentals collapse from heaven to hell, but three problems hitting at once: valuations were too full, rates were too stubborn, and expectations were too unrealistic.
We do not chase the loudest headline; we break down the most valuable logic. In this episode, we will make three things clear. First, did the broad AI semiconductor plunge mean the industry logic changed, or is Wall Street simply repricing AI trades at elevated levels? Second, were core names like Marvell and Micron hit because orders are weakening, or because they had already run too far and the market used macro rates and earnings gaps to slam valuations? Third, for ordinary investors, is AI semiconductors now a warning sign that the bull market is ending, or is it the next buy-the-dip window after weak hands have been shaken out?
Today we will explain the capital logic, rate pressure, and valuation reset behind this AI semiconductor bloodbath.
Let's first make clear the spark that ignited the selloff. On the surface, the market pointed the finger at Broadcom's earnings. But were Broadcom's results really bad? Not really. Broadcom's second-quarter revenue was about $22.2 billion, up 48% year over year, and AI semiconductor revenue was about $10.8 billion, up 143%. In any normal company, those numbers would let the CEO turn the earnings call into a victory party.
But in today's AI trade, the issue is not whether you are strong; it is whether you are strong enough to keep Wall Street euphoric. The market no longer wants a 90-point test score from AI semis. It wants 110 points, plus the next exam already finished in advance. Broadcom's problem was not that it was bad; it was that it was not surprising enough. Once a stock, a sector, and an industry chain are already fully owned by capital, being good is no longer enough. You have to be dazzling.
That is the first key logic today: the drop in AI semiconductors may not mean demand collapsed; it may simply mean the market cooled off after overheating expectations. It is like a hotpot restaurant with 500 people lining up outside. That means business is great. But if the landlord has already set rent as if 5,000 people queue every day, then even if the restaurant is still doing fine, the valuation may not hold.
Now look at the second trigger. U.S. May nonfarm payrolls were much stronger than expected, with 172,000 new jobs and unemployment holding at 4.3%. In a normal economic headline, that is resilience. In the eyes of tech-stock traders, that is the Fed-cut story getting kicked back out the door. Stronger employment means inflation and wage pressure are less likely to cool, and the market worries more that the Fed will not cut soon, or may even keep rates elevated for longer.
Why do high rates kill AI semis? Because AI semiconductors are not priced like traditional cyclical stocks. They are priced like super-growth stocks. Their share prices already embed several years, even a decade, of future growth imagination. The higher the rate, the less valuable those future cash flows are when discounted back. In plain language, investors once happily paid $100 for your dream of the future, but when rates rise, they suddenly say: hold on, I can earn yield in short-term bonds, shouldn't your dream get a discount?
So this was not only a Broadcom problem, and not only a Marvell or Micron problem. It was the entire AI chain running into the same issue: capital started recalculating prices. The market used to ask whether a company had AI. Now it asks whether the company has definite orders, gross margin, and free cash flow.
That is the stage where AI trading shifts from "storytelling at scale" to "verifying cash flow."
Let's talk about Marvell first. Marvell fell very hard, and many people immediately wondered whether the company had a problem. In fact, Marvell's fundamentals did not suddenly collapse. Its latest quarter set a revenue record, and its data-center business is a large part of the mix. AI interconnect, custom chips, and optical connectivity are truly in key positions within AI infrastructure.
Marvell's most compelling angle is not that it directly challenges Nvidia's GPUs, but that it sits at the AI data-center "connection toll booth." You can think of an AI data center as a super city: Nvidia is the most expensive brain inside it, Micron and Hynix are the memory system, TSMC is the factory, and what Marvell does often looks more like highways, bridges, fiber, and switching channels. The bigger the models, the more chips, the denser the servers, the more data needs to flow at high speed. If the data cannot move, even the most expensive GPU may become a group of rich people stuck in traffic inside a mansion.
That is why the market loved Marvell for a while: it was betting on a new bottleneck called interconnect. First everyone watched GPUs, then HBM memory became crucial, and now people realize that fast connections between different chips, racks, and data centers are just as critical. That made Marvell one of the representative stocks of the idea that AI is not just about buying GPUs.
But here is the problem: the more the market treats something as a bottleneck, the easier it is to price it with no room for error. Marvell's plunge did not mean AI interconnect was suddenly useless. It meant the market began to doubt whether the previous valuation had assumed three overly optimistic things: that AI data-center buildouts would never slow, that customers would never cut orders, and that Marvell could always maintain high share and high margins at key nodes.
If the market doubts even one of those assumptions, the stock will react first. Especially when a custom-chip giant like Broadcom reports and does not make the market cheer, capital starts to hit related names too. It thinks: if Broadcom cannot satisfy everyone, should Marvell, which had a stronger run and more concentrated expectations, also cool off first?
In short, Marvell's problem is not lack of value. The issue is that the market may have priced it as a "sure-win AI toll booth." And capital markets hate the words "sure-win." Once everyone thinks it is sure-win, the company often does not break first; the price does.
Now Micron. Micron dropping more than 13% scared people even more, because over the past few months it had become the face of the AI storage rally. HBM supply has been tight, DRAM prices have been rising, NAND and enterprise SSD demand have recovered, and data-center customers have been scrambling for memory.
The real change with Micron is that it is no longer just a traditional cyclical storage stock. In the past, DRAM and NAND were classic cyclical businesses: when prices rose, everyone talked about shortages; when prices fell, inventories piled up and prices collapsed faster than a mall clearance sale. But AI has lifted the strategic importance of storage. Especially HBM, which is not a normal memory stick but core support next to the AI accelerator. The stronger the GPU, the higher the demand for HBM. Nvidia, AMD, and other AI chip platforms all depend on high-bandwidth memory. That is one reason Micron got repriced by the market. Investors began to believe storage might no longer be just an old cycle, but a strategic resource in AI infrastructure. As a major U.S. storage supplier, Micron also gained extra scarcity value thanks to geopolitics, supply-chain security, and AI customer lock-in.
But why did it still plunge if the fundamentals were so good? The answer is simple: the old weaknesses of the storage industry never fully disappeared. If prices rise too fast, capex follows. If everyone expands production, the market starts worrying about oversupply down the road.
One of the things the market feared this time was whether future capacity expansion by Micron, Hynix, and Samsung would turn today's shortage into tomorrow's pressure. Especially when the stock price has already priced in very strong HBM demand, any news about supply increases, customer adjustments, or server-architecture changes can be magnified into a reason to crush valuations.
So was this bloodbath the end of the AI semiconductor rally? Please like and subscribe, and we will keep going.
I think it looks more like a crowded-trade stampede than a collapse in the industry trend. AI data-center construction is still underway, cloud giants' capex is still there, Nvidia GPU demand is still there, and the bottlenecks in HBM, interconnect, optical modules, power, and cooling are still there. But the stock market is not an industry white paper. Every day it must deal with funding cost, crowded positioning, valuation pressure, and sentiment swings.
One particularly important signal in this drop is that capital did not run away from equities as a whole. It pulled out of the most crowded AI semis. In other words, this was not end-of-the-world panic. It was a structural rebalancing. The names that had risen the most, traded the most expensively, and carried the highest expectations fell the hardest. Health care, financials, and defensive sectors held up relatively well. That tells us the market is not rejecting the economy; it is re-pricing the risk premium on tech stocks.
So some of you are wondering: after a drop like this, is it an opportunity or a trap?
From an investment-opportunity perspective, this AI semiconductor pullback does not mean the whole industry logic has failed, but it does raise the bar for stock selection and entry timing.
From a valuation perspective, the key is whether, after the selloff, expectations have returned to a more reasonable level. AI semiconductors have already rallied a lot, and some companies were indeed priced as if very high growth was guaranteed. After the pullback, we need to reassess whether current prices already reflect future order growth, margin improvement, and competitive structure, rather than just staring at the size of the decline.
Short-term trades are about volatility and timing; medium- and long-term trades are about industry trends and earnings realization. Those two must be judged separately.
If you care more about short-term trading, the semiconductor sector may see a technical rebound after a big drop, but that rebound depends mainly on sentiment, rate changes, and the speed of capital inflows.
If you care more about the medium and long term, then what matters after this plunge is whether the core competitiveness of the companies has changed. For example, are orders stable? Can cash flow support future R&D and expansion? As long as these core indicators have not deteriorated materially, the pullback is more about valuation pressure and sentiment release than a fundamental reversal.
After this selloff, the AI semiconductor sector is still investable, but opportunities will become much more differentiated. What is worth tracking is not every stock that fell, but those companies whose fundamentals remain solid, whose order visibility is high, whose valuation pressure has been meaningfully released, and whose position in AI infrastructure remains critical.
The first layer is the core AI infrastructure leaders. These companies may not be the cheapest, but they are the most certain. Think of firms controlling the GPU ecosystem, networking, custom chips, advanced packaging, and HBM supply chains. They will still move with valuation swings in the short run, but as long as customers do not materially cut capex, the long-term logic is not broken.
The second layer is the bottleneck names that were sold off too hard. For example, Micron in high-bandwidth memory, and Marvell in high-speed interconnect and custom chips. Their opportunity is not an immediate bounce the next day; it is that after panic subsides, if earnings continue to prove demand is real, pricing is firm, and margins are stable, valuation repair can happen quickly. But the prerequisite is data, not faith.
The third layer is the peripheral picks-and-shovels of AI infrastructure: power, cooling, optical modules, data-center equipment, and networking gear. These names often are not as sexy as GPUs, but as AI advances, the bottleneck is increasingly likely to shift from chips to power, connectivity, cooling, and space. AI is not a chip story; it is a full infrastructure rebuild. Smart capital often spreads from the hottest area to the scarcest area.
But I need to warn you of one thing: after a big drop like this, the biggest danger is not the market itself, but human emotion.
What we should remember from this AI semiconductor bloodbath is not just how much it fell, but that the market rules changed. Real AI opportunities will become more selective, more differentiated, and more demanding of investors' patience and discipline.
To sum it up in one sentence: the AI semiconductor bloodbath on June 5 wiped out not long-term AI demand, but crowded positions and overly expensive expectations.
Today's episode is not meant to scare everyone, and it is not meant to tell everyone to blindly buy the dip. It is meant to make clear that U.S. AI semiconductors have entered a harsher but higher-quality new stage. The old phase was story-driven. The next phase is earnings-driven. The old phase was everything rising together. The next phase is the strong surviving and the weak falling behind. Real investment value is not in the loudest shout, but in who can still present orders, profits, and cash flow after the big selloff.
