TSMC's 66% Margin, Intel Lands Tesla on 14A & Korean NPUs Hit Revenue

Jamie Cole (0:00)

AI Hardware & Chips, Daily News. I'm Jamie Cole, thanks for joining me. Today, TSMC's margin resilience can it hold through 2nm scaling.

TSMC just posted its strongest quarterly result on record, and the number that matters most isn't the revenue beat, it's the margin. 66% gross margin, sustained through the early phase of 2nm mass production.

That's the signal. Process transitions are where foundry margins typically compress. New nodes bring higher defect rates, slower yields and elevated capital costs. TSMC is doing it while holding the line on profitability. For now. The headline numbers back it up. Revenue came in at $46.72 billion, ahead of guidance. Net profit grew 68.7% year over year. Those figures reflect something structural, not cyclical. TSMC's fabs are running full, driven by sustained demand from Nvidia, AMD and Broadcom, all of whom are absorbing advanced capacity as fast as it comes online.

CC. Wei pointed to two demand drivers he sees extending the cycle, agentic AI deployment and broader open-source model adoption. The implication is that AI chip demand isn't narrowing to a handful of frontier labs, it's widening. Whether agentic workloads ultimately require the same silicon profile as foundation model training is still an open question. But the direction of demand is clear enough. Here is the part that deserves more attention than the headline suggests. TSCC's customer base is highly concentrated. Nvidia, Apple and Broadcom together control the majority of advanced node orders. And those customers set their spending plans 12 or more months in advance. The record quarter tells you what the last cycle looked like. It doesn't tell you what happens if one of those customers pulls back in 2026

Arizona fab costs are the other variable. Domestic US production carries a meaningful cost premium over Taiwan. As TSMC scales Arizona, that pressure will show up in the margin profile.

Watch the Q4 guidance closely for any signal on how management is pricing that in. Here's the thing. The other development worth holding alongside the TSMC story is Intel's confirmation of Tesla as a 14-anchor foundry customer. That matters because it removes the biggest single uncertainty around Intel's advanced logic roadmap, which was whether any major external customer would actually commit. Tesla committing to 14A is the first real external validation Intel has had for that node. It doesn't settle the yield and cost questions, those remain unconfirmed.

But it signals that sophisticated chip buyers are at least willing to run production on Intel's next-generation process, and that changes the competitive frame. Intel also announced a $20 billion equity raise earmarked for Foundry FAT buildout, covering 18A, 14A and advanced packaging. The raise finances the ambition. The harder test is whether it translates to margin-positive Foundry revenue before the decade ends. One more development that changes the longer-term picture. South Korean MPU makers Rebellions, FurosiaAI and DeepX have crossed from R&D into active revenue generation. Telecom operators in export contracts are now producing tangible income. AMD has backed at least one of these players. The signal here isn't that Nvidia is suddenly threatened. It's that the AI chip ecosystem is beginning to fragment in a way that creates real alternatives at the infrastructure layer, particularly for inference workloads, where power efficiency matters more than raw peak performance.

The real question this week sharpens to two things. Can TSMC sustain 66% gross margins into the deeper 2nm ramp as Arizona costs accumulate?

And is Tesla on 14A bring additional foundry customers to Intel? Or does it stay a one-customer validation? Those two data points will tell you more about the next 12 months in semiconductor infrastructure than almost anything else. Thanks for listening. This podcast was built using AI technology, a Yes We production.