TSMC printed record Q2 revenue on Thursday, and the stock fell 5% anyway. Taiwan Semiconductor beat across the board on July 16, 2026: revenue up 36% to NT$1.27 trillion, full-year growth now guided above 40%, and the capex budget lifted to $64 billion. The easy read wrote itself in seconds. A blowout got sold, so the AI trade must be tired. That read misses what the tape actually did. The drop wasn't the market rejecting the numbers. It was price resetting after a run-up that had already paid the earnings forward, and in wave terms that reset has a name.
The setup in 3 lines:
- TSMC beat every headline number and still dropped 5% on July 16.
- The sell-off is a "sell the news" reset after a pre-earnings run-up, not a demand break.
- Whether this is a buyable Wave-4 dip or a real top depends on the pre-earnings breakout holding.
Within an hour of the release, UBS analyst Crystal Hsu told clients the surprise capex hike "boosts confidence in the AI supply chain." Her desk read the same numbers as bullish. The stock kept right on selling anyway. So the gap wasn't about whether the quarter was good. It was about how much good news the price already carried before Hsu ever hit send.
TSMC's 36% revenue beat wasn't the number that moved the tape
Start with what TSMC actually reported. Revenue climbed 36% from a year ago to NT$1.27 trillion, a record, powered by AI orders from Nvidia, Apple, AMD and Broadcom. Management raised full-year growth guidance above 40%. The surprise came next: capex jumped to a $60 to $64 billion range, up from the prior $52 to $56 billion plan. TSMC almost never lifts capex in the second quarter, so a double-digit raise is a real tell about demand.
One number, though, cut the other way. The Q3 gross margin guide of 65% to 67% landed under the 67.5% some desks had penciled in. The culprit is the 2nm ramp, the newest and most advanced chip process, which trims margins by 2 to 3 points while the new fabs fill up. In plain terms, TSMC is spending hard now to build capacity that pays off later, and later is not this quarter.
That mix is why a beat could still sell. Bulls got the demand signal they wanted. Bears got the soft margin line they needed. Price had to pick a side, and for one session it picked down. The same tension showed up across the sector this week, right next to ASML's own $43 billion capex answer to the chip rout.
$64 billion. That's the 2026 capex line TSMC almost never touches in Q2, and it still jumped more than 10%.
The 5% drop is a sell-the-news reset, not a demand break
Price ran into this print already stretched. TSM had climbed for weeks on a June revenue figure that jumped 68% from a year earlier, and shares tagged fresh highs going into Thursday. When a name rallies that hard ahead of a known date, the news has to beat an expectation that price already moved to meet. That is the whole engine behind "buy the rumor, sell the news."
Think of it like a concert that sells out months early. The show can be great and the resale price still slips the next morning, because everyone who wanted in already paid up front. TSMC delivered the show, and the pre-earnings buyers had already paid.
So the 5% isn't the market saying AI demand rolled over. UBS, the capex line and the Nvidia order book all say the opposite. The 5% is the market letting air out of a move that got ahead of itself. And that difference is everything, because a demand break and a positioning reset build completely different structures from here. You can see the same split playing out in the AMD and Broadcom custom-silicon divergence.
Wave 4 is the pullback that shakes bulls out before the last push
Here is where structure earns its keep. In Elliott Wave analysis, a sustained uptrend usually moves in five legs: three pushes higher (waves 1, 3 and 5) split by two pullbacks (waves 2 and 4). Wave 4 is the second pause, and it lands right after the strongest, most obvious leg of the whole move. It is the dip that shows up when the story is loudest.
That timing is exactly why wave 4 fools people. The fundamental news sits at peak-bullish, record revenue and raised guidance, so a pullback feels wrong and reads as a top. Most bulls sell it and step aside. Then the fifth wave puts in the final high without them. The pullback did its one job, which is to move shares from impatient hands to patient ones.
Does the TSMC print fit that map? It lines up with it closely. The 68% June revenue run-up looks like the third-wave thrust, the loudest leg. A "good news gets sold" session right at peak sentiment is textbook fourth-wave behavior. Now, that is a read and not a certainty, and the tape still has to confirm it. A fourth wave that falls too far stops being a fourth wave, which is why the level it holds matters more than the headline that caused the dip.
Fibonacci turns the pullback into a level, not a mood
A wave label alone doesn't tell you where to care. Fibonacci does. Traders measure a healthy fourth-wave pullback as a share of the prior push, and the shallow retracements, roughly a third to just under half of the third wave, are the ones that keep the five-wave count alive. Put simply, a normal pause gives back some of the last rally, not most of it.
The reason this matters for TSM is that the pre-earnings breakout hands us a clean anchor. Price broke to new highs before the print, and that breakout shelf is the line a genuine fourth wave should defend. Hold above it on a weekly closing basis and the loudest-leg-then-pause reading stays intact. Lose it and stay below, and the count that says "one more high" is in real trouble.
A third to a half. That's how much of the prior rally a healthy fourth-wave pullback gives back before the last push, no more.
TSMC is one of the assets on the Kunkel Capital rotation, and members get the full structure map with the exact entry zone, exit target and invalidation level refreshed on a fixed cycle.
The $400 floor and the pre-earnings breakout are the levels to watch
Strip it down to what any chart shows. The pre-earnings breakout to new highs is the first reference, because that shelf is where a fourth wave either holds or fails. Below it sits the $400 round number, a level buyers have defended before and the kind of psychological floor that tends to draw bids on the way down. Those two lines, the breakout shelf and the round number, frame the zone where this call lives or dies.
Above the breakout, the job is simpler. If the fourth-wave read is right, the next real test is the pre-earnings high itself. Reclaim it and the fifth-wave case is back on the table, with the AI-capex demand signal doing the heavy lifting underneath. None of this needs a fresh headline. The numbers already came out, and from here it is a structure question, so structure resolves at levels, not in press releases.
Here is the honest part. This post gives you the shape: a record beat, a positioning reset, a likely fourth wave anchored to a visible breakout. What it doesn't hand you is the precise price to act on, because that comes off the wave count itself. The same discipline runs through how we mapped this week's soft-CPI bitcoin move and the bank earnings around the CPI double print.
Where this thesis breaks
Every structural read needs a line that kills it, and here is ours. The fourth-wave case breaks if TSM posts a weekly close back below the pre-earnings breakout while the AI-capex story reverses, think order cuts from Nvidia or a walk-back on that spending budget. Price slipping alone isn't enough. It is price and behavior together: the shelf gives way, and the demand signal that justified the run-up fades at the same time.
That is the Kunkel Capital signature. We always tell you how you would know the read is wrong, in terms you can watch on your own screen. The product tells you the exact price where wrong becomes official. For now, the two things to track are plain enough. Does the breakout shelf hold on a weekly close, and does the AI order book stay intact. The macro backdrop stays supportive too, the same soft-inflation read behind gold's real-yield squeeze this week.
The July 16 print in five points:
- TSMC beat with record revenue up 36% and lifted full-year growth guidance above 40%.
- The capex raise to $64 billion is the demand tell, well above the prior plan.
- The soft Q3 margin guide, tied to the 2nm ramp, gave bears their excuse to sell.
- A 5% drop after a run-up to new highs reads as a fourth-wave reset, not a demand break.
- The pre-earnings breakout and the $400 round number frame where the read holds or fails.
Frequently asked questions
Why did TSMC stock fall 5% after beating earnings on July 16, 2026?
The stock had rallied to new highs going into the report, so the good news was already priced in. A soft near-term margin guide, driven by the 2nm ramp, gave sellers a reason to take profits. That is a classic "sell the news" reset, not a sign that AI demand weakened.
Is the TSMC pullback a buying opportunity or the start of a top?
Structurally it looks like a fourth-wave pullback, the pause that lands right after the strongest leg of an uptrend. It stays a buyable dip as long as the pre-earnings breakout holds on a weekly closing basis. It flips bearish if that shelf gives way while AI orders slow.
What is a Wave 4 in Elliott Wave analysis?
A strong uptrend usually runs in five legs, with three pushes higher and two pullbacks. Wave 4 is the second pullback, and it tends to arrive when the news is most bullish, which is why it often shakes out impatient holders before the final push.
What would prove the TSMC bullish read wrong?
A weekly close back below the pre-earnings breakout, happening at the same time the AI-capex demand signal reverses. Price and behavior have to break together. A shallow dip that holds the breakout keeps the five-wave count alive.
Know your entry, your exit, and where you are wrong on TSMC
The 5% drop on a record beat is the surface signal. TSMC is on the Kunkel Capital watchlist: the full research maps the current wave count to a defined entry zone, an exit target and the exact invalidation level, refreshed on a fixed rotation, with alerts when levels hit. €19.99 first month, then €34.99. Cancel anytime.
Last updated: 2026-07-17
