Last night, U.S. stocks were brutal! The Nasdaq dropped 4.2%, and the Philadelphia Semiconductor Index plunged 10.3%. Chip stocks wiped out $1.3 trillion in a single day! And that wasn't even the scariest part! The most terrifying thing was that Wall Street actually went after the hottest, most profitable, and most heavily chased trade—the so-called "tech stock faith."
Over this entire leg of the U.S. rally, a lot of people only believed one story: AI was still accelerating, the tech giants were still pouring in money, data centers were still expanding, NVIDIA was still selling shovels, Broadcom was still making custom chips, Micron was still talking about AI storage, Marvell was still talking about interconnects and optical communications, and AMD was still talking about being the second supplier.
So whenever the market sold off, many funds instinctively thought, no worries—it will still come back to tech in the end.
But the huge red candle on June 5 was like a bucket of ice water dumped straight onto AI tech stocks.
NVIDIA fell, Broadcom fell, Micron fell, Marvell fell, and AMD fell too. The semiconductor sector was like it had its network cable ripped out all at once. The day before everyone was talking about AI's vast future; the next day Wall Street sat in the conference room and coldly asked, bro, isn't your valuation getting a bit too expensive?
Today, what we really need to break down are three things.
First, is this Nasdaq selloff the end of the AI bull market, or just a high-level shakeout in tech stocks?
Second, why did tech stocks fall harder than the broader market, and why did semiconductors become the epicenter?
Third, if the market repairs next, who will money buy first, and which AI infrastructure theme will it keep digging into?
I'll put the conclusion up front.
This tech selloff is not the end of the AI story. It is the second phase of the AI trade.
In phase one, the market bought imagination. As long as a company could talk AI, talk compute, talk data centers, funds were willing to give it a high valuation.
In phase two, the market buys execution. Do you have orders? Do you have revenue? Do you have profit? Do you have cash flow? Do you have real pricing power?
In short, AI is not over—but the era of blindly buying tech stocks is over.
When many people saw the plunge, their first reaction was panic: is the tech bull market over? Others got excited and thought, time to buy the dip—moving their hands before their brains, with accounts more daring than their wallets.
But mature investors do not react with panic or excitement first. They dissect the logic.
Is the thing falling valuation, or fundamentals? Is the thing falling sentiment, or trend?
If it is only sentiment and valuation, the drop may create an opportunity. If fundamentals have turned, then the rebound is not a buy signal—it may just be a chance to leave with dignity.
So let's start with the first blade: rates.
The Nasdaq is not a normal index. It is fundamentally a growth-stock index, a future-cash-flow index. Tech valuations are not about how much you make today; they are about how much you may make many years from now.
So who decides what future money is worth?
The answer is interest rates.
The lower the rate, the more valuable future money is. The higher the rate, the less valuable future money becomes.
That sounds like finance class, but it is easy to understand. Suppose an AI company tells you it will make a lot of money ten years from now. If market rates are very low, investors say, fine—I am willing to pay a high valuation now, because discounted back, that future money still looks attractive.
But if rates rise, capital changes its face immediately. A moment ago you were AI's shining star; now they ask, if risk-free yields are this high, why should I pay such a rich price for profits that may only arrive in ten years?
That is why tech stocks fear rates so much.
The key change in the market was that U.S. nonfarm payrolls for May added 172,000 jobs, well above expectations, while unemployment stayed at 4.3%. When that data came out, the market's first reaction was not joy—it was nerves.
Because stronger jobs mean a more resilient U.S. economy, and the Fed has less reason to rush cuts. In fact, the market may start debating whether higher rates need to stay higher for longer.
For ordinary people, strong employment sounds good. But for high-valuation tech stocks, strong jobs can feel like a bitter coffee: it wakes you up, but it is bitter too.
So the first blade that cut the Nasdaq was not AI demand—it was the valuation multiple.
Remember this: tech stocks do not usually die first from bad news. They often die first from being too expensive.
The second blade is overcrowded positioning.
For a while, the U.S. market has not been rising because all stocks rose together. It has been rising because a small group of tech giants and AI semis pulled the index upward.
NVIDIA was the king, Broadcom represented custom chips and AI networking, Micron represented AI storage, Marvell represented AI interconnect and optical-communication expectations, and AMD represented the second-supplier theme.
These stocks had already absorbed too much capital, too much emotion, and too much imagination.
When they rise, everyone thinks this is the main theme, and anyone not onboard is missing out.
When they fall, the problem is that everyone bought the same logic, so when they run for the exit, they use the same door.
It's like a wildly popular restaurant with lines stretching three blocks. Everyone says the place represents the future of dining. Then someone yells, “This meal isn't as amazing as advertised.” The scariest part is not that the food is bad—the scariest part is that the line is too long, and everyone is afraid of being the last one out, so before anyone has even eaten, there is already a stampede at the door.
AI tech stocks are like that now.
The story did not vanish in one day. It is just that there were too many people at the front and the entrance was too narrow. When sentiment turns, the selling gets amplified.
The third blade was Broadcom igniting the mood.
Broadcom is not a bad company at all. In fact, it is one of the most important companies in AI infrastructure. It has custom chips, networking chips, large cloud customers, and long-term demand from AI data-center expansion.
Broadcom's second-quarter AI semiconductor revenue reached $10.8 billion, up 143% year over year. In any ordinary industry, that is not merely good—it is absurd.
If you are not just here to watch the fireworks, but really want to understand the money flow behind this tech selloff, go ahead and like and save this. Over the next few days the market probably won't be gentle; it may give you an opportunity while also testing your heart. If it were gentle, I'd almost suspect Wall Street took the day off.
Once a stock is priced as an AI core asset, a merely decent report is no longer enough. It has to deliver a report so good that Wall Street raises expectations again.
That is the cruel part of AI stocks now. Good is not enough. Growth is not enough. Profit is not enough.
You must prove you can keep making money faster in the future—and faster than Wall Street's already extreme imagination.
So what Broadcom really triggered was not one company's issue, but the valuation issue across the whole AI trade.
The market started asking whether AI chip demand had already been fully priced in, whether cloud capex would slow, when AI data-center ROI would arrive, whether custom chips would take share from GPU budgets, and whether rising financing costs would squeeze upstream orders.
These questions always existed, but in a bull market no one wanted to hear them.
When stocks are going up, everyone talks about the stars and the sea. When they are falling, everyone starts checking the footnotes in the earnings report. In a bull market, a PPT can be read like poetry. In a crash, even the cash-flow statement gets reviewed like a crime scene.
So what kind of drop is this?
I divide tech selloffs into three types.
First, a sentiment crash. Fundamentals are not clearly worse, but the market is panicking and selling first. For example, ETF redemptions, quant deleveraging, institutions cutting risk, and many stocks getting sold indiscriminately.
Second, a valuation crash. The company is still growing and the industry is fine, but the stock was too expensive before, and the market no longer wants to pay that multiple. In other words, the company is fine, but the stock was too rich.
Third, a fundamental reversal. Orders decline, revenue slows, gross margins fall, customers cut orders, inventory rises, and the cycle turns. This is the most dangerous, because it is not a drop that creates opportunity—it means the trend really changed.
This time, I lean toward sentiment plus valuation. There is still no evidence that AI infrastructure has entered a fundamental reversal.
Why do I say that?
Let's look at a few hard numbers.
NVIDIA's latest quarterly revenue was $81.6 billion, up 85% year over year, with data-center revenue of $75.2 billion, up 92%. What does that tell us? It tells us AI compute demand has not suddenly disappeared, and cloud providers and enterprise customers are still building AI infrastructure at scale.
Broadcom's issue is not lack of growth—it is that the market wants even more explosive growth.
For memory companies like Micron, the logic does not disappear just because the Nasdaq drops for one day. AI training needs HBM, inference needs wider memory bandwidth, and data centers need enterprise SSDs and high-capacity storage. AI does not just consume GPUs; it also consumes memory, storage, networking, power, and cooling.
The market is smashing stock prices, not pulling AI servers out of data centers.
So the most accurate read is this: the AI fundamentals have not broken immediately, but the valuation tolerance for AI stocks has become much lower.
What does lower valuation tolerance mean?
In the past, as long as a company said AI, the market was willing to forgive weak profits, weak cash flow, and expensive valuations. People thought the future was huge, so paying a little extra now was fine.
Now it is different. In a high-rate environment, capital becomes picky. You either bring certainty, bring upside surprise, or bring real cash flow. Otherwise the market will not keep paying a story premium.
So AI is not unbuyable—it just can't be bought the old way anymore.
In the past, buying AI meant buying the hype. Now buying AI means buying execution. In the past, buying tech meant seeing who had the biggest story. Now buying tech means seeing who can turn the story into profit.
That means tech stocks will diverge sharply from here.
First category: core leaders.
For example NVIDIA, Microsoft, and Broadcom. They have real customers, real revenue, real profit, and real cash flow. They are the foundation of the tech-stock faith. If these names fall, the market will first see whether they can stabilize, because they decide whether the AI main theme is still intact.
NVIDIA represents standardized AI compute. Microsoft represents cloud and AI application deployment. Broadcom represents custom chips and AI networking. If all three cannot stabilize, then the market is not just correcting—it is cooling the entire tech valuation framework.
Second category: high-beta AI infrastructure names.
For example Micron, Marvell, and AMD. They are not the steadiest leaders, but if AI capex keeps broadening, they may become the next wave of money flow.
Micron represents AI storage. AI needs more than GPUs—it needs HBM, DRAM, NAND, and SSDs. The larger the model, the more inference there is, and the more data there is, the more important storage and memory become.
You can think of GPUs as AI's brain, HBM and DRAM as AI's short-term memory, and NAND and SSDs as AI's grain silo. No matter how smart the brain is, if memory is not enough and the silo is not enough, it cannot run.
Marvell represents AI interconnect and data-center networking. Future AI will not be one chip fighting alone; it will be tens of thousands of chips forming a giant compute factory. How the chips connect, how data moves, how latency is reduced, and how power consumption is controlled—these are the opportunities for companies like Marvell.
AMD represents the second-supplier logic. The market buys AMD not because it can beat NVIDIA today, but because AI compute demand is so large that cloud providers cannot rely on one supplier forever. As long as AMD captures some share, it has upside.
Third category: pure-story tech stocks.
These are the most dangerous. They have AI in the name, big-model slides in the PPT, robots on the launch stage, and CEOs speaking like a sci-fi movie trailer—but no revenue conversion, no profit conversion, and no cash-flow conversion. These are the names most likely to get de-rated during the reality check.
Because the market no longer wants to pay for fuzzy futures.
Looking back at last night’s selloff, it is not enough to say tech is dead, nor is it enough to say buy the dip. What we really need to do is reclassify tech stocks.
Who is a core asset? Who is a beta asset? Who is just a sentiment asset?
Core assets can wait for stabilization after the panic flushes out. Beta assets should wait for leaders to stabilize before money rotates down the chain. Sentiment assets need extreme caution because they may be the next big de-rating target.
Let's answer another key question: why don't I lightly call the AI main theme over?
Because the long-term logic of AI infrastructure has not been broken.
First, cloud giants have not stopped spending capex.
Microsoft, Amazon, Alphabet, and Meta keep pouring money into data centers because they know future cloud computing, search, advertising, enterprise software, office systems, content generation, customer service systems, and security systems will all be reshaped by AI.
This is not a one-quarter story. It is a ten-year infrastructure migration.
Second, AI demand is not a single-point demand—it is a system demand.
Whenever many people talk about AI, they only stare at NVIDIA. But an AI data center does not just buy GPUs. It needs GPUs, CPUs, memory, storage, switches, optical modules, power, liquid cooling, land, racks, and networking.
So even if the market eventually disagrees with GPU valuations, the AI infrastructure chain will not vanish. Capital will simply search within the chain for the tightest, most priced-in, and most profitable segment.
Third, the shift from training to inference will create a new demand structure.
A few years ago, AI discussions focused on training large models. Training is the biggest GPU consumer, which is why NVIDIA became the absolute core. But as AI applications roll out, inference demand will get much bigger. Inference is not one-off; it happens every day, every hour, every minute.
Every time you ask an AI a question, generate a video, or call an enterprise agent, there is inference cost underneath. The more inference there is, the higher the requirements for chip efficiency, memory bandwidth, network connectivity, and storage read/write performance.
That means AI infrastructure is not a one-time buildout; it is continuous consumption, continuous upgrading, and continuous expansion.
That is why the AI main theme still cannot be easily turned bearish.
But I want to stress one thing: not turning bearish easily does not mean charging in blindly now.
True mature trading is not yelling “opportunity” just because stocks drop, and it is not shouting “collapse” either. It is waiting for confirmation signals.
First signal: can the Nasdaq stop falling?
If the Nasdaq keeps making new lows, it means risk appetite for tech has not returned. At that point, don't rush to call a bottom, because the index is telling you sentiment is still unstable. A healthy repair is not a one-day bounce; it is several days without new lows, plus renewed participation from leading tech names.
Second signal: can the Philadelphia Semiconductor Index recover?
Semiconductors were the hardest-hit area in this selloff and are the most important thermometer for the AI hardware chain. If the Philly Semiconductor Index keeps falling, it means capital is still cutting AI-hardware valuations. If it stabilizes first and rebounds stronger than the Nasdaq, then money is coming back to AI infrastructure.
Third signal: can NVIDIA hold up?
NVIDIA is the kingpin of the AI trade. If it cannot stabilize, don't trust rebounds in other AI names too easily. A healthy rebound starts with NVIDIA stabilizing first, Broadcom following, and then Micron and Marvell spreading the move.
Fourth signal: can Broadcom stop the bleeding?
Broadcom represents custom AI chips and networking. If it keeps breaking down, confidence in the second layer of AI infrastructure will remain pressured. If it stabilizes and the market reaffirms its AI semiconductor growth, that means the market is not rejecting AI—it is re-screening valuations.
Fifth signal: what happens to U.S. Treasury yields?
If Treasury yields keep rising, tech stocks will have a hard time rallying comfortably. The higher the rate, the harder it is for tech valuations to expand. If yields cool off, or at least stop breaking higher, then tech has room to re-rate.
Sixth signal: are cloud providers cutting AI capex?
The biggest fear for the AI upstream chain is not a one-day stock drop—it is customer order cuts. As long as Microsoft, Amazon, Alphabet, and Meta keep confirming AI data-center spending, the logic for chips, memory, networking, power, and cooling remains intact. But if those giants clearly slow AI capex, then this is no longer just a valuation issue—it becomes a fundamental risk.
So what matters next is not one stock's daily move, but the linkage.
If the Nasdaq stops making new lows, the Philly Semiconductor Index starts to recover, NVIDIA stabilizes first, Broadcom stops falling, Treasury yields stop rising, and cloud capex is not cut, then this selloff is probably not the end of the AI bull market. It is more likely a high-level rotation within the AI infrastructure theme.
How does money move then?
Step one: money buys the leaders first.
After panic, institutions do not first buy small caps—they return to the most certain assets. NVIDIA, Microsoft, and Broadcom are the first names money watches and tests.
Step two: money buys beta.
If leaders stabilize, money starts looking for laggards to catch up. At that point Micron, Marvell, and AMD—the high-beta AI infrastructure names—will re-enter focus.
Step three: money buys the derivative plays.
If the AI theme is confirmed to be repairing, sidelined money starts looking for what can rise besides NVIDIA. At that point AI storage, AI interconnect, data-center power, liquid cooling, servers, and optical communications will be rediscovered.
That is the usual diffusion path in a bull market: first the hardest-hitting leaders rise, then the strongest second-tier names, then the spillover themes.
What we need to do is not guess the bottom, but see whether money is really returning to the main theme.
Final takeaway.
This selloff is not telling us AI is finished. It is telling us AI stocks have entered the verification phase.
In the past, you bought the story. Now you buy execution. In the past, you watched who was loudest. Now you watch who has real orders, real profit, and real cash flow.
If it is only a sentiment crash, the leaders will come back first. If it is only a valuation crash, good companies will get capital again at better prices. But if fundamentals have turned, the rebound is a trap.
So in the days ahead, watch these key names.
Whether NVIDIA can stabilize AI faith. Whether Broadcom can stabilize the custom-chip and AI-networking logic. Whether Micron can stabilize the AI-storage logic. Whether Marvell can stabilize the AI-interconnect logic. Whether AMD can prove the second-supplier theme is still alive.
Then watch whether the Nasdaq and the Philadelphia Semiconductor Index can recover in sync.
If those signals all appear together, then this tech plunge may not be the end of the AI bull market—it may be a reshuffling for the next AI infrastructure wave.
But if none of those signals show up, the market keeps breaking support, yields keep rising, and the tech leaders keep selling off, then don't rush into heroic dip-buying. The market won't hand you the answer just because you want to make money.
Tech stocks are not dead, but the era of buying tech stocks blindly is over.
In the next phase, the companies that survive will not be the loudest storytellers, but the ones that turn AI into orders, profits, and cash flow.
The huge red candle on June 5 was not just a drop—it was also a test paper.
It tests not whether you dare buy the dip, but whether you can tell the difference between trend, sentiment, valuation bubble, and true long-term opportunity.
