This AI Chip Stock Just Signed Massive Deals With 3 Hyperscalers, and It Still Looks Like a Great Buy Right Now (Hint: Not Nvidia or Intel) - finance.yahoo.com

Skyler Monroe (0:10)

Hey, everyone, welcome back to the AI Hardware Show. I'm your host Skyler Monroe. And if you're into chips, silicon, and the hardware that's literally powering the future of intelligence, you are in exactly the right place. Before we dive in, huge thanks to our sponsors. Ada, helping businesses integrate AI into their real world workflows. Ago Consulting, that's Ago, your go-to for silicon development from AI accelerators to full SOZE design. And Zen Semiconductor, the umbrella company building the processors and AI fabric behind modern data centers. Think Sierra RISC-V CPUs, Loom AI Fabric, and Ada for deploying AI at scale. These folks are doing serious work, and we're grateful to have them. All right, on today's show, we've got a mystery AI chip company locking up deals with three major hyperscalers. JP Omergen throwing some cold water on the AI chip rally. Apple possibly going big on AI servers? A detailed investigative report on how Nvidia chips are sneaking into China despite export controls? And OpenAI's custom chip called Jalapeño is apparently turning up the heat on Nvidia's margins. Let's get into it. OK, story one, and this one comes with a little mystery attached to it. The headline is, this AI chip stock just signed massive deals with three hyperscalers, and it still looks like a great buy right now.

And the big clue?

It's not Nvidia.

It's not Intel. So who are we talking about?

Well the reporting strongly points to Broadcom, which has been quietly becoming one of the most important players in the custom AI silicon space.

And the reason this matters so much is the three hyperscalers part. When we say hyperscalers, we're talking about the giants, think Google, Meta, Amazon, Microsoft. These are companies running infrastructure at a scale most of us can barely comprehend. Thousands of servers, millions of users, petabytes of data moving around every second.

So when a chip company locks in deals with three of them at once, that's not a coincidence, that's a strategy paying off. Broadcom's angle here is custom AI accelerators, sometimes called XPUs or ASICs, Application-Specific Integrated Circuits. Now here's a good way to think about the difference. Nvidia's GPUs are like a Swiss army knife, incredibly versatile, powerful across a huge range of tasks.

But an ASIC is more like a custom chef's knife, designed specifically for one job, and in that job, it absolutely excels. Hyperscalers love ASICs because they can tailor the chip precisely to their workloads, whether that's training large language models, running inference, or processing recommendation algorithms. And Broadcom is really good at helping these companies design and manufacture those custom chips. Now the finance angle here is interesting too. The stock is getting attention because even after the AI boom has driven valuations up across the board, analysts are still saying this one has room to run.

That's a signal that the market believes custom silicon is not a trend, it's a structural shift, and I think they're right. We are moving into an era where one-size-fits-all compute is giving way to purpose-built silicon.

The hyperscalers figured that out years ago, and Broadcom was smart enough to position themselves right in the middle of it. This is a story worth watching closely, because the more hyperscalers invest in custom chips, the more the entire landscape of AI hardware shifts. Alright, story two, and this one is a little more of a reality check.

JPMorgan is out with a note saying the AI chip rally could slow down, and the reason they give is that hyperscalers are closing the performance gap with their own custom silicon. So this is kind of the flip side of the story we just talked about. If hyperscalers get good enough at building their own chips, and they are investing billions to do exactly that, do they still need to buy as many GPS from Nvidia? That's the core question here, and it's one the market is starting to take seriously. Now, let's put some context around this. Nvidia's dominance in AI compute has been pretty extraordinary. Their H100 and B200 GPS became the gold standard for training and running large AI models. And the demand has been so intense that at times these chips were essentially on backorder for months. Companies were paying massive premiums just to get their hands on them. But here's the thing, that kind of pricing power and demand only holds if there isn't a good enough alternative. And JPMorgan is suggesting we're approaching a point where the alternatives are getting real. Google's TPS, Tensor Processing Units, are a great example.

They've been in development for years, and each generation has gotten meaningfully better. Amazon has Tranium and Inferentia. Microsoft is developing their own AI chips, too. These aren't science projects anymore. They're production hardware running at massive scale. Think of it like this. Imagine you're a restaurant that used to buy all your bread from one bakery because it was the only place making artisan loaves. But now three or four bakeries in your neighborhood have figured out the recipe.