Samsung Building Facility With 50,000 Nvidia GPUs To Automate Chip Manufacturing

An anonymous reader quotes a report from CNBC: Korean semiconductor giant Samsung said Thursday that it plans to buy and deploy a cluster of 50,000 Nvidia graphics processing units to improve its chip manufacturing for mobile devices and robots. The 50,000 Nvidia GPUs will be used to create a facility Samsung is calling an "AI Megafactory." Samsung didn't provide details about when the facility would be built. It's the latest splashy partnership for Nvidia, whose chips remain essential for building and deploying advanced artificial intelligence. [...]

On Thursday, Nvidia representatives said they will work with Samsung to adapt the Korean company's chipmaking lithography platform to work with Nvidia's GPUs. That process will results in 20 times better performance for Samsung, the Nvidia representatives said. Samsung will also use Nvidia's simulation software called Omniverse. Known for its mobile phones, Samsung also said it would use the Nvidia chips to run its own AI models for its devices. In addition to being a partner and customer, Samsung is also a key supplier for Nvidia. Samsung makes the kind of high-performance memory Nvidia uses in large quantities, alongside its AI chips, called high bandwidth memory. Samsung said it will work with Nvidia to tweak its HBM4 memory for use in AI chips.

womens shoes

[noshellswill]

Fifty-thousand graphics cards in use ... without a product specified. Really? Perhaps the graphics cards will be used as heating elements ( blocked-chained heaters ?) for two-hundred guild craftsmen making hi-end womens shoes. I know of no product with documented buying frenzy and unreasonably higher markup than women shoes ... and they can put the factory on a street called "ITALY", so they can advertize crafted Italian womens shoes. To say nothing of the product being generated by family

More like Zoho

[kurt_cordial]

I assume you're C-Suite? A 420x69p resell-lution requires substantiation.

Re:

[DeadBeef]

It's the generic all purpose business plan these days:

1. Announce in media you have signed up purchase order to buy a giant pile of nvidia jank.
2. ???
3. Profit!

Market Dependencies.

[geekmux]

It's the generic all purpose business plan these days:

1. Announce in media you have signed up purchase order to buy a giant pile of nvidia jank. 2. ??? 3. Profit!

If the S&P is now vitally dependent on the apt-corruptly named Magnificent Seven, don't be surprised if most of AI becomes wholly dependent on nVidia, who seems to be bound and determined to become the next TSMC. Which a planet now relies on.

Would Greed abuse a single stock to crash entire markets if it could (hedge) a Wall Street bet against such activity and win? Guess that depends on how legal that shitty behavior is.

Re:

[haruchai]

all that's needed now if for Elon to announce that Grok has designed a GPU that's an order of magnitude more efficient than Nvidia's

Re:

[haruchai]

called it!
https://electrek.co/2025/11/03... [electrek.co]

Overlords

[ketomax]

So the GPUs are now making more GPUs. The singularity just went self-sufficient.

Re:

[geekmux]

So the GPUs are now making more GPUs. The singularity just went self-sufficient.

Just gonna say it now. First AI CEO will be the nVidia CEO.

Unless..

Farts, All the Way Down

[crunchy_one]

It's part of a circular investment scheme, and they're not even hiding it. Samsung provides Nvidia with chips from their fabs and Nvidia sells them back to Samsung. In other words, they're huffing each others farts in order to tart up their respective books. The coming crash is going to be breath taking.

Re:

[gweihir]

The coming crash is going to be breath taking.

Indeed. And yet, the usual idiots will learn nothing and the usual assholes will just try again in a while.

Hallucinations on silicon?

[gweihir]

Tell me more about this bad idea! Will be entertaining to see.

Why?

[boxless]

How exactly do 50 thousand gpus gel Samsung make better chips? Is it something used during production, or design?

They talk about using simulation software. Is that it? The article is unclear beyond that little tidbit.

Re:

[ffkom]

The article indeed is incredible vague about everything, and the claim "20 times better performance" is nonsensical when there is not even the slightest hint what metric is meant with "performance" there.

Re:

[unrtst]

That phrase, "20 times better performance," drove me to the comments to see if anyone had any clue what that was about :-)
I still don't know, but I read it as, "Samsung will make GPUs of Nvidia's design in their facility; Their fab is better, so it will result in 20 times better performance out of the same design," even though I think that'd be damn near impossible. They also mentioned hooking up new memory, so maybe it's the result of all the improvements together? Beats me.

Re: Why?

[Gavino]

It works like this. C-suite buy a bunch of nvidia shares. Then they come out with some big positive nvidia announcement. If the nvidia stock price goes up, they sell half their nvidia shares cashing in, but still hoping there will be more hype in future that sends the shares higher. If the stock price doesn't move, then they know that the bubble has peaked and sell all before the bloodbath. It's just how hype trains work. Mostly market manipulation through smoke and mirrors. Dot com boom bust was exactly th

Re:

[tlhIngan]

Well, there are some areas of manufacturing that are somewhat unpredictable.

First, performance variation - you can have a chip design, and have a huge variation in performance across the wafer and you need to characterize this. This is often done using test wafers specifically mis-manufactured to be exceptionally slow and exceptionally fast and performing simulations and tests to verify the logic is still functioning at either extreme.

Second - yield improvement - flaws in chips still remain and is one of th

Re:

[Admiral Trigger Happ]

They are way behind TSMC terms of yield quality on the latest nodes, so they need to be doing something. But the bottleneck is ASML's EUV lithography hardware.

Galaxy AI

[Samare]

I guess Samsung will stop using Gemini AI for its Galaxy AI.

AI for nVidia's sake?

[geekmux]

I plan on starting an AI company that just sits around and talks about AI companies.

Therefore I need at least eleventy-bazillion nVidia GPUs and three nuclear power plants worth of juice.

It's the latest splashy partnership for Nvidia..

Yeah. How "splashy" is that justification? Are we talking ware just barely wet with vapor or what?

Re:

[sphealey]

"I plan on starting an AI company that just sits around and talks about AI companies.

Therefore I need at least eleventy-bazillion nVidia GPUs and three nuclear power plants worth of juice."

I am prepared to invest $2 billion into your startup. Please advise where to wire funds. Thanks you oh so much for this ground floor opportunity! /s

I think I can translate this

[packrat0x]

The GPU cluster will be used for an AI that will control the CPUs used in the design and/or manufacturing process.

The Worm Ouroboros

[sphealey]

At some point people are going to realize that Ouroboros or a mobius strip takes you nowhere and just costs you money to keep circling forever.

(maybe the smart Wall Street money already has. Hmmm....)

They don't have a clue

[Farley1]

They don't know how 50,000 GPU chips is going to actually help them. They have to build it first, and then ask it.

Re:

[Ostracus]

The answer is 42.