Meanwhile, Michael Burry is down bad… 

As we all know, the chip sector spent six weeks getting measured for a casket, and then the only company that actually knows what's coming posted the best month of its life. TSMC said Monday that July revenue hit NT$467.58 billion, or about $14.5 billion, up 44.7% from a year ago and 5.6% from June. (Yes, that's a month, you dorks…  not a quarter.)

In short, TSMC's monthly revenue drop is the closest thing the market has to a live AI demand tracker, because nobody in this trade builds their own chips. Nvidia and Google design theirs; neither one manufactures anything. They hand the blueprints to the same company on the same island, which owns about 72% of the global foundry market (Samsung has 6, and yes, that's the whole race). There are maybe thirty buildings on Earth that can play The Odyssey the way Nolan shot it, and people drove past four multiplexes anyway. Leading-edge silicon works the same… but with one address.

Case in point: Q2 revenue was $40.2 billion, up 36%. Net income jumped 77%. Gross margin hit 67.7%, which is girthy gainz for a business whose factories cost north of $20 billion apiece. High-performance computing, where TSMC books its AI silicon, made up 66% of revenue. Management guided full-year growth above 40% and raised capex to $60–64 billion. That 44.7% puts July ahead of the company's own 40% target, which Ben Barringer at Quilter Cheviot called "no mean feat" before immediately warning that people "do not read too much into the monthly numbers as they can jump around." (Congrats on the record. Now don't use it for anything.) He's not wrong. One month is one month, chip demand turns on a dime, and TSMC doesn't comment on these releases at all.

Friendly reminder, the chip trade has been getting the body-bag treatment since June. The PHLX Semiconductor Index sits about 15% below its high (still up 72% on the year, so hold the eulogy), chip names briefly fell into a bear market, and ASML got knee-capped 6% in one session on a report that China started mass-producing its own DUV lithography machines. Monday it clawed back 2%, with Infineon and STMicro along for the ride. Nobody in that selloff was arguing AI demand is fake. They were arguing over who eats sh*t on the bill. Turns out it isn't the fab.

C.C. "Robust" Wei owns exactly one adjective. AI demand is "extremely robust" today, and on the Q2 call he said it stays that way through 2030. Of course, he’s not a poet… he is, however, mid-pour on $265 billion in Arizona, the largest foreign direct investment in American history and a spectacularly expensive way to be wrong.

Which brings me to the reason Michael Burry is probably reading this article right now, the bubble argument. In the simplest of terms, it’s a fight over whether hyperscalers ever earn a return on the ~$730 billion they're spending this year. PIMCO figures that eats roughly 94% of their operating cash flow. TSMC sits on the other side of that bet entirely. Sub-5nm prices went up 3 to 5% this year (read: the fourth straight year of telling Jensen the price went up), 2nm wafers run about $30,000, and the book is full into 2028. Their ROI question is this company's revenue line.

“Short it”, they say. Call the top, tell your group chat it's f*cking over, screenshot the SOX chart. Meanwhile, you’ll still be standing in the same line on the same island behind Apple, holding a purchase order for $30,000 wafers that don't ship for two years, and the bubble will have raised prices on you on the way in. There’s no “wei” out. Until next time, friends… 

At the time of publishing, Stocks.News holds positions in Apple and Google as mentioned in the article.