For six straight quarters, the TSMC trade has been the easiest read in tech. AI demand goes up. TSMC’s profit goes up. Repeat.

Last Wednesday, the pattern held again, and then some. Revenue beat guidance. Net income jumped 69% year over year. CEO C.C. Wei stood up on the call and called AI demand “endless.”

Then the stock fell anyway.

Not a little. TSM dropped as much as 5.3% intraday before closing down 1.8%, and it kept sliding into Friday.

Not because the quarter was bad. Because of what sits underneath it.

TSMC is now more dependent on one customer group than at any point in its history, right as the AI debt story everyone’s been whispering about started showing real cracks. This is the first week that story showed up directly inside TSMC’s own numbers.

Today I want to walk through what the market actually flinched at, because it is not the number you’d guess.

Missed Friday’s Playbook on ASML?

Tesla finally unveiled the long-delayed next-generation Roadster, complete with SpaceX-developed thrusters, sub-one-second 0-to-60 mph claims, and even limited airborne stunts. The catch is timing: production is not expected until 2028, years later than originally promised. The bigger takeaway is that this is more of a halo product for Tesla’s brand and engineering image than a major near-term earnings driver.

Google Cloud could post more than 100% year-over-year revenue growth when Alphabet reports on Oct. 28, helped by explosive AI demand and a backlog that recently reached around $514 billion. The important part is not just growth, but profitability: Cloud is becoming one of Alphabet’s biggest sources of operating-income expansion. In plain English, Google is increasingly looking like an AI infrastructure company as well as a search giant.

Retail investors have been taking profits, but institutions may be stepping in just as analyst sentiment improves. Institutional ownership is still only around 45% to 46%, leaving room for more fund buying, while recent upgrades from Goldman Sachs, Barclays, and others point to growing Wall Street confidence. The next big test is Q3 earnings, especially whether Palantir can back up Alex Karp’s aggressive free-cash-flow targets and keep sovereign AI momentum going.

Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:

After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.

The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.

“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”

Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.

Learn More

Taiwan Semiconductor posted $46.7 billion in quarterly revenue last week, up roughly 41% from a year ago.

Net income hit $23.6 billion, up 69%. Sixth record profit quarter in a row.

The obvious read: AI chip demand is unstoppable, and TSM is the one company that gets paid no matter who wins the AI race.

That read is not wrong. It is just incomplete.

The stock sold off on the news anyway, and the number that explains why is buried three slides into the earnings deck, not in the headline.

Here is what I found when I went looking for it.

Revenue of $46.7 billion beat TSMC’s own guidance range and came in roughly 2% ahead of Wall Street’s estimate, according to LSEG data.

Net income of $23.6 billion was up 69% year over year and 8% from the prior quarter. That is not deceleration. That is acceleration into a record.

High performance computing, Wall Street’s shorthand for AI and data center chips, now makes up 68% of TSMC’s revenue. A year ago it was 57%.

That is an eleven-point swing in twelve months. That is the fastest reshaping of a trillion-dollar company’s revenue mix I have tracked in this job.

C.C. Wei told analysts demand still exceeds what TSMC can supply. He used the word “endless.”

Guidance backs him up. TSMC is pointing Q4 revenue toward $52.3 billion to $54.3 billion, roughly 14% above this quarter at the midpoint.

And the next leg is already shipping. TSMC’s newest and most expensive node, 2 nanometer, jumped from 3% of wafer revenue to 8% in a single quarter.

Here’s what sits underneath that 68% number.

A year ago, TSMC’s revenue leaned on more than one leg: smartphones, PCs, automotive, and AI, each one real on its own.

Now 68% of revenue sits on top of a single theme: AI infrastructure spending.

That spending has a funding problem the market is only now pricing in.

This week, CoreWeave’s bonds traded like a company under real stress, with yields pushing past 16% and credit default swaps spiking to record levels. A London hedge fund mapped $3.6 trillion in AI financing deals across the sector and flagged more than 120 of them as circular, meaning the same companies show up as each other’s investors, customers, and suppliers at once.

TSMC does not carry that debt. Its customers increasingly do.

Nvidia has conditionally committed close to $100 billion to OpenAI. OpenAI has committed hundreds of billions to the cloud providers building the data centers that will eventually need TSMC’s chips.

If any link in that chain slows its spending, TSMC’s fastest growing, highest margin business is the one standing closest to the blast radius.

The margins already show a hint of that strain. Gross margin slipped to 66.4% from 67.7% last quarter, and TSMC’s own Q4 guidance points a touch lower again, mostly from the cost of ramping 2nm production.

That is a small number by itself. But it is the first margin contraction of this entire AI cycle, and it arrived in the same quarter TSMC became more dependent on AI spending than ever.

TSM spent most of the back half of this year climbing steadily toward its 52-week high of $487, and heading into earnings it still looked like a stock that could do nothing wrong.

The reaction to a genuinely great quarter should have been another leg higher. Instead, TSM gapped down hard on earnings day, touched $407 intraday, and closed out the week around $422, roughly 5% below both its 50-day and 200-day moving averages.

This is what happens when good news runs into a market that already decided to be nervous about something else, in this case the same AI financing jitters rattling CoreWeave and the broader chip complex.

The $407 earnings-day low is acting as the first line of defense for now, with a tougher test waiting near $391 if that level breaks.

On the upside, TSM needs to reclaim somewhere around $438 to $440, where its moving averages are stacked, before the chart starts looking like the old uptrend again.

Right now the stock is stuck between a very good earnings report and a very nervous tape, and which one wins out from here says a lot about whether this AI cycle still has the market’s trust.

The CoreWeave situation is the wildcard sitting over this entire sector right now.

Bull case: Nvidia keeps insisting the financing is sound, and this panic fades within a few weeks the way past AI scares have.

Bear case: more AI infrastructure names show the same stress, spending plans get trimmed, and TSMC’s order book takes the first real hit of this cycle.

Wildcard: credit markets, not fundamentals, end up deciding how this plays out, and TSMC’s stock gets dragged around by headlines that have nothing to do with its own chips.

2nm already tripled its share of wafer revenue in a single quarter, and every additional point of mix shift is pure margin upside once the line matures.

Bear case: new nodes run expensive before they turn profitable, and if margins keep slipping while 2nm scales, the market starts questioning whether “record profit” is quietly masking a less efficient business underneath.

Wildcard: TSMC never discloses yield rates in detail, so how smoothly this transition actually goes is something we only see in the margin line, after the fact.

TSMC is weighing new capacity in Texas beyond its existing Arizona build-out, and reportedly listening to Elon Musk’s pitch for his Terafab project.

Bull case: more US capacity chips away at the geopolitical discount investors have priced into TSM for years.

Bear case: US fabs run at lower margins than Taiwan fabs, and a bigger US footprint could become the next drag on the margin story, not a fix for it.

Wildcard: how much of this expansion ends up shaped by Washington policy rather than TSMC’s own roadmap.

Watch whether HPC and AI revenue keeps climbing past 68% or starts to plateau.

I covered TSMC back in October 2024 at $186 in TSMC Up 83% YTD: Is This Your Last Chance To Ride The Chip Boom, arguing the AI story still had real room to run.

The stock has more than doubled since then. The setup has changed completely. Concentration is the risk now, not conviction.

Bull case: TSMC has historically leaned into strength by raising its dividend, and a buyback signal here would tell you management sees this selloff as noise, not a warning.

Bear case: no change in capital return policy, paired with continued heavy capex guidance, would suggest management is bracing for a longer AI buildout cycle than investors want to sit through.

Wildcard: any fresh comment on US tariff exposure, which has hovered in the background of every TSMC call this year without ever fully resolving.

Here’s where I’ve landed after sitting with this for a few days.

TSMC did not have a bad quarter. It had one of its best ever, and the stock still fell.

That tells me the market is not actually questioning TSMC’s execution. It is questioning everyone standing behind TSMC in line.

Markets rarely punish the supplier when the real problem sits with the buyer.

Eventually they figure out the difference, and the stocks that kept delivering while the panic played out are usually the ones that get re-rated higher once it clears.

I think TSMC is one of those stocks.

And honestly? The riskiest thing you could do here is confuse a financing scare in CoreWeave’s balance sheet for a demand problem in TSMC’s order book.

Those are not the same thing, and right now the chart is handing you a discount for the market’s confusion.

I am not selling this dip. I am watching $407, and treating anything close to it as a gift.

If you have paid time left on an individual Granola plan, we'll match it with a Wispr Flow subscription that includes Notetaker and dictation, and add bonus time, up to 12 months total. Sign in or create a Wispr account and submit proof of your plan to check eligibility.

Try Notetaker Free

The best lesson from this week isn’t about chips.

It’s about learning to read what a stock is actually reacting to.

A company’s earnings and its stock price in the same week are not always measuring the same thing.

Sometimes the chart is pricing in your company’s future. This week, TSMC’s chart was pricing in someone else’s balance sheet.

Separating “my company has a problem” from “my company’s customers might have a problem” is one of the hardest skills an investor can build.

Get that distinction right often enough, and you will spend a lot less of your life panic selling good businesses on bad days.