News Intel and TSMC agree to form chipmaking joint venture: Report

[phead128]

This makes sense for TSMC and the US if a calamity is expected to hit Tawain. It might make sense for INTC investors if they believe TSMC expertise will fix Intel fab problems.

Chips have to be sent back to Taiwan for advanced packaging, so unless you re-create the entire supply chain at scale back in US, including advanced packaging, then the world is going to take a hit no matter what.

 

[jp7189]

Chips have to be sent back to Taiwan for advanced packaging, so unless you re-create the entire supply chain at scale back in US, including advanced packaging, then the world is going to take a hit no matter what.

As far as I know, Intel doesn't package in Taiwan. They do the opposite, fab in Taiwan, ship elsewhere for packaging. They have packing in New Mexico and Arizona, and I seem to recall hearing something in Malaysia. TSMC has plans for packaging in the US... not sure of the timeline.

 

[3en88]

Who benefits how much is a function of power. Deals like this would not have happened without Trump. This is a full transfer of tech from TSMC to intel. Intel will make all of its tiles in house. TSMC's 20% stake in a paper company without assets is so that they feel like they get something too. Asians are very sensitive to "face" so optics are important. What they really get is a promise from the Trump administration to be left alive. They know the jig is up. The US government can destroy TSMC at any moment and now the US government is finally using its power.

 

[bit_user]

Chips have to be sent back to Taiwan for advanced packaging,

I'm pretty sure Intel has its own packaging plants. Where do they package Lunar Lake and Arrow Lake? Or Granite Rapids, for that matter?

so unless you re-create the entire supply chain at scale back in US, including advanced packaging, then the world is going to take a hit no matter what.

TSMC is currently planning on adding advanced packaging capacity in the US. I'm not sure how soon that'll be online.

 

[bit_user]

This is a full transfer of tech from TSMC to intel.

Not in the sense that Intel can just take that IP and run with it. Obviously, some information will leak out about how TSMC does things that could theoretically benefit Intel, but I expect a lot about how the lithography machines are programmed & the theoretical underpinnings will be closely guarded by TSMC. There's no way they'd give away the crown jewels, like that.

What they really get is a promise from the Trump administration to be left alive. They know the jig is up. The US government can destroy TSMC at any moment and now the US government is finally using its power.

Not sure exactly what you're implying, but no.

 

[thestryker]

That will never make any sense to me.
A FAB can not have more than 100% capacity, if you have 100% capacity with your own products you get to sell them at a premium, you are not just paid to manufacture them you are also getting paid the normal margin.

Here's the thing: Intel has never had a problem with margins and fab operation. The problem they have is that by being their own supplier and needing to keep them running at capacity they retire nodes quickly to make sure they don't. They would basically cycle a generation into chipsets and other lower margin parts before retiring them. While that didn't used to be a problem it's completely untenable for EUV nodes.

It's going on 6 years since Zen 2 launched and N7 based nodes are still making up a good chunk of TSMC revenue. It took 2 years after release before N5 revenue passed N7 and N3 didn't pass N7 until 2024. I couldn't quickly find full year 2024 data but N3 was up and down all last year which indicates it still hasn't caught up to N5. Consider that N2 will be full volume next year this should speak volumes as to the issues at hand with EUV node costs.

To keep nodes around that means Intel needs more fab space, and to pay for that they need more customers so they can keep the new space running at capacity.

 

[thestryker]

With the GPUs, I think I read somewhere that Intel 7 wasn't well-suited to GPUs. Not sure about why they didn't opt for Intel 3, but perhaps cost?

I think a lot of the Intel 3 avoidance comes from capacity and delays due Intel 4 problems than anything else. Remember GNR was supposed to be an Intel 4 part originally, but it seems pretty clear the node wasn't up to par even though they made a lot of MTL chips with it.

TSMC is currently planning on adding advanced packaging capacity in the US. I'm not sure how soon that'll be online.

Hasn't even broken ground yet.