Nvidia just got punched in the face by the market last week. $1.8 trillion in market cap, gone. The Nasdaq dropped 4.18%. The bears were having a field day — "peak GPU" this, "custom chip" that, "the capex cycle is cooked." They were dancing on NVDA's grave before the body was even cold.

Then Apple walked into WWDC and blew the whole thing up.

Here's what happened: Apple announced its revamped Siri runs on Nvidia Blackwell B200 chips through Google Cloud. Not Apple silicon. Not some homegrown neural engine. Nvidia. The company that refused to touch Nvidia GPUs for a decade. The company with the best silicon team on the planet — the crew that built M-series chips so good they made Intel look like a relic. That Apple just rented Nvidia's hardware to power its flagship AI product.

The Admission Nobody's Talking About

Let this sink in. Apple has $600 billion in annual revenue. Trillion-dollar R&D budgets. The M4 chip is a marvel. And they still couldn't build silicon that competes with Blackwell for large-scale AI inference.

They didn't partner with Nvidia because it's convenient. They partnered because there is no alternative.

This is Apple's biggest third-party hardware dependency since it put Intel inside the Mac in 2005. And the scale is bigger. This isn't about laptop CPUs — it's about the AI backbone of a platform with 2 billion active devices. Every Siri query from every iPhone, iPad, Mac, and Vision Pro is about to hit a Blackwell GPU.

The Custom-Chip Thesis Just Died

The entire bear case against Nvidia rests on one idea: hyperscalers will ditch Nvidia GPUs and build their own chips. Google's got TPUs. Amazon's got Trainium. Meta's got MTIA. Microsoft is supposedly cooking something up. The bears have been screaming for two years that Nvidia's moat is temporary — that custom silicon would commoditize the GPU market the same way generic drugs commoditize pharma.

Apple just proved that thesis is fantasy.

If the richest company on Earth with the most talented chip designers in the world can't build competitive AI silicon, who can? Google TPU? Please. Amazon Trainium? Cute. Those chips work for internal workloads. They're not threatening Nvidia's ecosystem. They never were.

The $1.8 Trillion Gift

NVDA is still down roughly 15% from last week's rout. The fundamentals didn't change. Demand for Blackwell is still insane. The capex cycle isn't slowing — hyperscalers are spending $300 billion this year. And now a $600 billion gorilla just validated Nvidia's moat in the most public way possible.

Apple chose Nvidia because Nvidia is the only game in town. Not for training. For inference — the part of the market the bears said would get commoditized first.

The stock got punished on macro fear and Broadcom's guidance whiff. That's noise. The signal is Apple running Siri on Blackwell with encrypted computing so Google Cloud can't even see the data. That's a moat wrapped in a moat. Confidential compute on Nvidia hardware is so far ahead of anything custom silicon can offer that Apple — Apple! — didn't even try to compete.

The Bottom Line

The bears said Nvidia's party was over. Apple just ordered another round for the whole bar. Blackwell demand was already through the roof before WWDC. Now it's got the world's most valuable consumer platform as a reference customer.

This isn't a warning. It's a generational buy-the-dip setup. When the dust settles on the AI selloff, NVDA is going higher — and the custom-chip thesis is going in the trash where it belongs.

— The Signal