Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/nvidia-intel-plan-to-rely-on-tsmc-in-fabricating-revolutionary-chips-for-data-center-pcs.23656/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Nvidia, Intel plan to rely on TSMC in fabricating 'revolutionary' chips for data center, PCs

Fred Chen

Moderator
Sep. 18, 2025 2:21 PM ET
By: Brandon Evans, SA News Editor

Nvidia (NASDAQ:NVDA) CEO Jensen Huang said Taiwan Semiconductor Manufacturing (NYSE:TSM) will provide the foundry support for the "revolutionary" chips they are making with Intel (NASDAQ:INTC).

"Nvidia and Intel are both successful customers of TSMC," Huang said during a press conference on Thursday afternoon. "They are a world-class foundry and support customers of diverse needs. You can't overstate the magic that is TSMC. But today, our partnership is 100% focused on the custom CPUs we are building for data centers that can connect to the Nvidia AI ecosystem."

"We both still have a lot of respect for TSMC, and we will continue to work with them," said Intel CEO Lip-Bu Tan. "We will still work on 14a and 18a and see if that can be used at some point in the future."

As part of the deal, which featured a $5B investment by Nvidia in Intel, the two companies will use Nvidia's NVLink, bringing Nvidia's artificial intelligence and accelerated computing strength and Intel's x86 architecture. They plan to combine Intel CPUs with Nvidia GPUs to create products for data center and PC markets.

"Together, our companies will create custom CPUs for data center platforms," Huang said. "For personal computing, we are going to create new Nvidia chips that integrate and fuse the world's best CPUs and GPUs. The era of accelerated and AI computing has arrived. Intel and Nvidia are partnered to drive it forward."

"This corroboration will build on the core strengths of both companies," Tan said. "This will unleash the new era of x86 innovation. I'm excited about what we can create together. Clearly, this is all about the scale. We can do even more working together ... Right now we are focused on collaboration, and we can talk about the process down the road."

Huang said the new products will have a total addressable market of more than $50B when considering their utilization in data centers and PCs.

"There is an entire segment of the market that ties in CPU and GPU that Nvidia has largely not addressed," Huang said. "We use NVLink to fuse the CPU and GPU into a new class of integrated laptops the market has not seen yet ... This is going to address some $50B per year opportunity ... The data center CPU market is $25B and the PC notebook market is 150M sold per year."

Huang also said the deal will not affect its relationship with Arm (ARM).

"We will continue to build our Arm roadmap, and this won't affect that at all," he said.

He said the deal was created solely by Huang and Tan.

"The Trump Administration had no involvement in this partnership at all," Huang said. "They would have been supportive, of course. I told (U.S. Commerce) Secretary (Howard) Lutnick today, and he was very excited about two American companies working together."

Analysts believe the partnership provides upside for Intel, but the company still needs to show more that it's back as a chipmaking power.

"This investment in INTC represents a large vote of confidence from NVDA," said Julian Lin, Investing Group Leader for Best of Breed Growth Stocks. "The momentum may go a long way towards the company's turnaround, but I emphasize that it still needs to show more signs of progress towards that turnaround. In comparison to other stories like Oracle (ORCL), this latest development appears to have more hopes built into the momentum."

"A bit of an alliance between NVDA and INTC should be a net negative for AMD (AMD), as it means more powerful competition, I believe," added Jonathan Weber, Investing Group Leader for Cash Flow Club. "TSM may lose some foundry business to Intel in the future but remains very well positioned and the clear manufacturing leader for the time being."

Arm shares had slipped 3.5% during afternoon trading, while TSM was up 2%. AMD had declined 1.5%.

https://seekingalpha.com/news/44962...ating-revolutionary-chips-for-data-center-pcs
 
By the way, it seems that TSMC does not package die made by other foundries together with die made in-house.
Basically, it seems that they only package die made in-house.
 
Xeon dies are not moving to TSMC as they’re the Intel fab volume, so what is being discussed here? That Nvidia will keep building their GPU dies at TSMC? That’s obvious after the press conference.

It’s TBD what the chiplet architecture looks like, but it’s possible Intel could fab the NVLink stuff too.

By the way, it seems that TSMC does not package die made by other foundries together with die made in-house.
Basically, it seems that they only package die made in-house.
It was made very clear in the press conference that Intel advanced packaging is what is enabling this TSMC+Intel chiplet combos.
 
By the way, it seems that TSMC does not package die made by other foundries together with die made in-house.
Basically, it seems that they only package die made in-house.
that is correct for logic die. Obviously they put Hynix packaged HBM products on the Cowos unit.
TSMC also subcons part of the COWOS process to other companies.
 
Xeon dies are not moving to TSMC as they’re the Intel fab volume, so what is being discussed here? That Nvidia will keep building their GPU dies at TSMC? That’s obvious after the press conference.

It’s TBD what the chiplet architecture looks like, but it’s possible Intel could fab the NVLink stuff too.
exactly it's poorly worded by the article
 
Xeon dies are not moving to TSMC as they’re the Intel fab volume, so what is being discussed here? That Nvidia will keep building their GPU dies at TSMC? That’s obvious after the press conference.

It’s TBD what the chiplet architecture looks like, but it’s possible Intel could fab the NVLink stuff too.


It was made very clear in the press conference that Intel advanced packaging is what is enabling this TSMC+Intel chiplet combos.
On the Datacenter side, It looks like Intel will fab the Custom Xeons at IFS with support for NVLink and sell it to Nvidia. Nvidia buys that and sells it to the ODMs (like Pegatron, Wistron or Foxconn etc) along with their. The ODMs using their suppliers puts the GPUs & the CPUs in a Bianca board with BGA soldered connection, assembles a rack scale NVL72/144 server and sells it to the Dell, HPs of the world or directly to cloud customers (not sure about this). This is incremental server CPU volume for Intel, so it benefits both Intel Products & Intel Foundry as incremental revenue.
Nvidia gets a x86 version of NVL72/144 to compete with AMD's upcoming Helios (x86 + AI GPU) rack scale solution. Some people were saying Grace CPU is not adequately working out for the NVL72 & Vera CPU (future custom ARM product) might fix that. This x86 partnership with Intel could be a hedge by Nvidia in case Vera flops.

On the Client side, it looks like Nvidia will fab the RTX iGPU chiplet at TSMC and supply that to Intel. Intel may fab their CPU at TSMC or IFS depending on the product but it will 100% package this CPU & RTX chiplet at IFS and sells it to the Notebook ODMS to make the laptop PCB boards and make the notebooks. I don't know who pays for the SoC packaging here, may be both Intel & Nvidia shares the cost of packaging at IFS. I don't think this is huge benefit for IFS unless this leads to sale of more laptop chips (taking markets share back from AMD). Otherwise it is pretty neutral or may be beneficial on the edges (meaning minimal) imo. For Nvidia this is really beneficial as they get to play in the market they don't right now (laptop iGPUs).

In the end, I do think both will be benefit from this deal. Intel gets liquidity to reduce the debt on the balance sheet (or whatever LBT wants to do with this, Intel has $17B extra cash this Q, what is he planning to do with this will be interesting, i hope it is not just debt reduction?), gets to play in one of the fast growing AI server segment and incremental revenue. Nvidia gets to play in iGPU market and gets a very good ROI for their investment in Intel shares.
 
That statement seems to contradict the Tom's Hardware article which said the two companies has had secret co-development projects going on for more than a year.

There were indeed preliminary agreements with PG but no 5 billion.
 
Back
Top