For years, Nvidia has been the dealer everyone in AI had to call. You wanted to train a model? Nvidia. You wanted to run inference at scale? Nvidia. You wanted to feel the existential dread of a supply chain bottleneck? Nvidia, baby, all day long.

But according to a report from TechCrunch, that cozy monopoly might be entering its villain arc - because practically every major tech player is now cooking up their own silicon.

Meet Jalapeño - yes, really

OpenAI, the company that somehow convinced the world that a chatbot deserves a $157 billion valuation, just revealed plans for its own custom inference chip. It's called Jalapeño. It was built in collaboration with Broadcom, and yes, it is named after a pepper. Whether this represents bold branding or someone just really hungry during a naming meeting is unclear, but either way, it signals something big: OpenAI wants off the Nvidia drip.

And they are absolutely not alone in this.

The gang's all here

Google has its TPUs. Apple has its M-series and Neural Engine silicon. And now SpaceX - yes, the rocket people - is apparently also in the custom chip business. When Elon Musk's rocket company starts designing AI accelerators, you know something structural is shifting in the industry.

The common thread here isn't just about performance or cost, though those matter enormously. It's about something every supply chain manager has nightmares about: single-supplier risk. When one company controls the hardware your entire business runs on, you are one bad earnings call, one export restriction, or one factory fire away from a very bad quarter.

So is Nvidia actually in trouble?

Let's not write the eulogy just yet. Nvidia still dominates training workloads, still has CUDA - the software ecosystem that every AI researcher has basically grown up inside - and still posts numbers that would make a medieval king blush.

But the inference market, where models actually run in the real world at scale, is a different game. It's higher volume, more cost-sensitive, and increasingly something that big companies would rather own themselves than rent from a single vendor at a premium.

The Jalapeño chip is specifically an inference chip. That's not a coincidence.

The bottom line

What's happening right now is less a revolution and more a slow, expensive, very deliberate divorce from dependency. These chips take years and billions of dollars to develop. Nobody is unplugging their Nvidia clusters tomorrow.

But the direction of travel is clear. The AI chip market is fragmenting, and Nvidia - for the first time in a while - has genuine reason to look over its shoulder.

Jalapeño is just the appetizer.