AMD Explores Samsung Foundry Deal as AI Chip Demand Drives Supply Expansion - The420.in

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 AI, you're in the right place. Big thanks to our sponsors today, Ada, helping businesses integrate AI into their workflows seamlessly. Ago Consulting, that's Ago, your go-to for silicon development consulting from AI to SOE. And Zen Semiconductor, building the processors and AI fabric behind modern data centers. Think Sierra RISC-V CPUs, the Loom AI fabric, and a whole portfolio of silicon industry ventures, plus Ada for deploying AI at scale. Today we've got a packed show, AMD is making moves with Samsung, JPMorgan is pumping the brakes on the chip rally, Cerebras is taking on Nvidia in the inference wars, someone's predicting the biggest AI chip winner of 2027 And we get hands on with Nvidia's GB300 in a desk side form factor. Let's get into it. All right, kicking things off, AMD is reportedly exploring a foundry deal with Samsung. And this one has some pretty significant implications for the AI chip landscape. So let's break down what's actually going on here. Right now, AMD, like most of the major fabless chip designers, relies heavily on TSMC for manufacturing.

TSMC is basically the gold standard.

They've got the best nodes, the best yields, and they've been the backbone of chips from Apple, Nvidia, AMD, you name it. But here's the thing. When everyone is rushing to that same well at the same time, capacity gets tight fast, and that's exactly what's happening. The demand for AI chips has been absolutely relentless. Every hyperscaler, every startup, every enterprise is trying to get their hands on more silicon, and TSMC's fabs, as impressive as they are, can only produce so much. So AMD exploring Samsung as an additional foundry partner makes a lot of strategic sense when you think about it that way.

It's about supply diversification, pure and simple. Think of it like this. Imagine you run a restaurant, and you have one incredible supplier for your most important ingredient.

Business is booming, orders are through the roof, and suddenly your supplier says, hey, we're pretty backed up. It'll be a few more months.

That's a nightmare scenario.

So you start looking for a second supplier, even if they're not quite as good, just to make sure you can keep the kitchen running. That's AMD's situation right now.

Samsung's foundry business has had some well-documented struggles with yield rates, and catching up to TSMC's leading edge nodes, that's not a secret in the industry. But Samsung has been investing heavily in improving their process technology, and from AMD's perspective, even having access to Samsung's capacity for certain product lines could free up TSMC capacity for their most critical, highest-performance chips. Now, it's worth noting this is still in the exploration phase. We're not talking about a signed deal or a major production commitment. But the signal here is important. It tells you that AMD is thinking seriously about scaling supply to meet what they clearly believe will be sustained. Long-term demand for AI hardware. And honestly, if you look at AMD's trajectory with their MI Series AI accelerators, the MI300X has been genuinely competitive. They have real reason to want more manufacturing firepower. This is a story worth watching closely.

Okay, shifting gears. JPMorgan has dropped a note that's getting a lot of attention in the investment community. And it's a bit of a cold shower for anyone who's been riding the AI chip wave with pure optimism. The headline is that JPMorgan thinks the AI chip rally could slow down. And the reason they're citing is that hyperscalers, the big cloud players like Google, Amazon, Microsoft and Meta are closing the performance gap with their own custom silicon. Let me unpack that because there's a lot going on here. For the past few years, Nvidia has had this almost unchallengeable position in AI training and inference. Their H100s, their H200s, these things have been the default answer to what do I use to train a large language model? The performance was just in a different league compared to alternatives. And that meant massive, massive demand flowing straight to Nvidia and by extension to AMD and other chipmakers trying to catch up. But here's what JPMorgan is pointing to. Google has TPS, which have been quietly getting very good. Amazon has their Tranium and Inferentia chips. Microsoft is working on their Maya AI Accelerator. Meta has their MTIA chips. These companies are not just dabbling in custom silicon anymore. They are making serious multi-billion dollar bets on it. And when the biggest buyers of AI chips start making their own chips, that changes the demand equation in a fundamental way. Think about it like this.