This week the market convinced itself the AI capex boom was cracking.

CoreWeave’s bonds started trading like distressed debt. Dell got downgraded for showing no “meaningful AI inflection.” Taiwan Semi beat every number on the board and still got sold off on margin worries.

Hot inflation data gave the selloff its excuse. The real argument underneath it was always about whether AI spending itself was overbuilt.

Then ASML reported.

Record sales. Guidance raised for the third time this year.

Order interest stretching out to 2030. Supply sold out through 2028.

The stock dropped anyway.

If the AI buildout were running out of room, the company selling the machines would feel it first. It just said the opposite.

That is the real story of this week, not the panic. The gap between what the machines are telling you and what the price action is telling you.

Today’s Pragmatic Playbook digs into why the company that builds the actual machines behind every advanced chip just said the AI buildout is real, and why Wall Street shrugged it off.

Missed Wednesday’s deep dive on SanDisk?

CoreWeave’s credit-default swaps have blown out above 1,100 basis points, implying roughly a 64% probability of default within five years, while some of its bonds briefly yielded around 16%. The bigger takeaway is that investors are starting to question whether debt-funded AI infrastructure can generate cash flow fast enough to support the borrowing. In plain English, the AI boom is still alive, but the financing side is suddenly looking much less comfortable.

Apple’s $1,999 iPhone Duo could become a strong premium upgrade product, but the Fool argues it is unlikely to be a game changer by itself. Estimates suggest 6 million to 10 million units in early sales, which would help revenue but still represent a relatively small slice of Apple’s overall business. The more interesting long-term angle is whether the Duo strengthens Apple’s broader ecosystem and gets existing users to spend more.

Tesla’s Terafab ambition is classic Musk: strategically logical, technically brutal, and potentially very expensive. Owning chip design, fabrication, and packaging could give Tesla tighter control over FSD, robotaxis, and Optimus, but advanced semiconductor manufacturing requires expertise in yields, process control, and clean-room operations that Tesla does not yet have. With capex already above $25 billion and the stock trading above 300 times earnings, investors will want proof this is disciplined vertical integration rather than another open-ended capital project.

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

This week gave you three different warnings about AI spending.

CoreWeave’s bonds traded like distressed debt. Dell got cut to Equal Weight for showing no real AI inflection. Taiwan Semi posted a record quarter and still watched its stock fall on margin worries.

The easy read is that the AI buildout is finally running out of room.

ASML makes that read harder to believe.

The company that builds the lithography machines behind basically every advanced chip on earth just posted record sales, raised its own guidance for the third time this year, and said customers are already talking about orders through 2030.

Its stock fell anyway.

That gap, between what the equipment maker is actually seeing and what the market decided to do with the stock, is what The Pragmatic Playbook is here to pull apart.

ASML’s Q3 wasn’t close.

€11.7 billion in net sales. €5.0 billion in net income. A margin beat on top of the revenue beat.

Full-year guidance got raised for the third time in 2026 alone.

Management said something more pointed than the usual earnings-call hedge. Customers are already discussing order volume through 2030, years past the current supply runway.

Supply is sold out through 2028.

Not “demand looks healthy.” Sold out.

The machine behind the machines just told you the AI buildout has years of committed spending ahead of it, not months.

Contrast that with CoreWeave’s bonds trading at distressed levels this week on fears the AI infrastructure boom is over-leveraged. Contrast it with Dell getting downgraded because Barclays could not find a “meaningful” AI inflection in its numbers.

ASML sits one layer below all of that noise, closer to the actual decision to build or not build the next generation of chip capacity. That layer just said yes, emphatically, for years.

Three guidance raises in a single year is not a normal pattern for a company this size.

Here is the part the bulls do not want to sit with.

The stock fell close to 4% the day ASML reported all of that good news.

Not despite the beat. Because of what came after it.

“Priced in” is a lazy explanation most weeks. This week it might actually be true.

ASML has spent 2026 rallying on exactly this story, that AI chip demand would keep climbing and orders would keep flowing. When a thesis gets confirmed in full, there is often nothing left to do but sell the ticket you already bought.

That is the uncomfortable lesson sitting inside Friday’s price action. Being right about the fundamentals and being right about the stock are two different trades, and this week ASML proved it is possible to nail the first one and still lose the second.

There is a second layer here, and it is less obvious than the headline beat.

ASML’s China sales reportedly surged more than 100% year over year this quarter, even with US export restrictions still in place.

That sounds unambiguously bullish until you look at the share, not just the dollar figure.

China’s share of ASML’s total sales actually declined from a year ago.

Not because China demand collapsed. Because everywhere else grew even faster.

Still, a healthier problem than the alternative. It just does not settle the bigger question hanging over the China relationship.

Export rules can tighten again. Beijing’s own chip ambitions can shift the order book again. Nothing about that relationship is a straight line just because this quarter’s number looked strong.

ASML has been here before, and it adjusted without the business breaking last time.

ASML has spent most of the year climbing with the rest of the AI trade, and heading into this week’s print the stock was sitting near the top of its range, around the $1,780 mark on the ADR.

Then the earnings hit, and instead of extending the move, the stock gave back close to 4% in a single session.

It tells you more about positioning than about the business. A lot of good news was already baked into that price before the report even came out, and once the number matched the story, there was nobody left who needed to buy.

The level worth watching heading into next week sits around $1,500. Some desks think the stock could test support there if the post-earnings drift continues.

Whether buyers show up at that level will say a lot about whether this is a one-day valuation reset or the start of something uglier.

For now the stock is still up sharply on the year, and the fundamentals behind it did not get weaker this week. They got stronger. The price just has not decided whether it believes that yet.

Here is what carries into next week.

ASML talked about order interest through 2030. The bull case is that this becomes a formal backlog number investors can underwrite. The bear case is that “interest” quietly becomes “intent” and then quietly becomes nothing, the way soft commitments sometimes do. The wildcard is a competitor or customer confirming the same multi-year demand independently, which would turn ASML’s claim from a talking point into a fact.

Watch for any fresh signal out of Washington or Beijing on chip equipment export rules. A tightening would hit the China revenue line directly. A loosening would remove one of the only real overhangs left on the stock. I covered the chip-capex math question from a different angle in Marvell’s $90 billion bet, and the same test applies here.

If CoreWeave’s debt troubles are really about one over-levered balance sheet, this stays contained. If they are a preview of how the whole AI financing stack is built, ASML’s order book will eventually have to answer for it too, no matter how full it looks today.

The stock did not fall because the business got worse. It fell because the price had nowhere left to go on good news alone. Next week’s test is whether ASML needs to beat again just to stand still, or whether this was a one-time reset.

TSM’s own margin outlook spooked the market this week even as its profit jumped on the same AI demand ASML is describing. If TSM’s capacity plans firm up from here, that is new demand flowing straight back into ASML’s order book.

I keep coming back to the same thought this week.

The market spent five days scared of a story, and one day ignoring the company best positioned to confirm or deny it.

ASML did not hedge. It raised guidance for the third time this year and said the order book stretches to 2030.

That is not what a company staring down a demand cliff sounds like.

Markets do not always reward being right immediately. Sometimes they punish it, because being right was already the consensus and there was no surprise left to pay for.

That does not make ASML’s print less true. It makes the stock’s reaction a pricing problem, not a business problem, and those two things resolve very differently over time.

I would rather own the company whose order book says 2030 than the one whose bondholders are pricing default risk. The stock market will eventually agree with me. It just has not gotten there yet.

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

This week tried to teach you that AI spending is cracking.

One company that actually builds the machines said otherwise, in writing, with numbers attached. That ought to count for something, even in a week built on noise.

The lesson for the weekend is not “buy ASML.” It is that the loudest story of the week is not always the truest one.

Debt scares and downgrades make noise. Order books sold out to 2028 do not need to shout.

Spend the weekend asking which one you have been reacting to.

Come Monday, try to react to the order book.