We've known that Apple has moved onto 5nm for a while now.
AMD securing more 7nm capacity has little to do with overcoming chip shortages.

The chip shortages are caused by insufficient supply of ABF substrates (which was caused by the pandemic and automotive industries mainly).

Without ABF substrates, TSMC cannot make new chips for AMD (no matter how much 7nm capacity they have at their disposal)... and this also applies for everyone else (Samsung, etc.).

Also, 7nm capacity was never an issue for AMD.
It was reported that TSMC always had enough capacity to meet AMD's needs (in the past, AMD underestimated the demand for its 7nm chips), but more recently, the insufficient supply of ABF substrates caused industry-wide chip shortages.

However, AMD DID recently agree to acquire Xilinx later this year.
Xilinx actually produce their own Integrated circuits... and one of the main reasons they agreed to this deal was to help AMD get more wins in the data-center space (where big money is), and I suspect that if they can do that, then its quite likely they will might also be able to supply TSMC with ABF substrates to for the entire AMD line of CPU's and GPU's.

Not true. AMD is less than 10% of TSMC business and Apple is more than 25%. Apple and AMD are both on TSMC 5nm today and will be on TSMC 3nm at the end of 2021 and 2022.

Samsung 3nm (GAA) is less dense than TSMC 3nm and has yield problems so no Samsung will not be competitive with TSMC at 3nm.

Here is the proper 3nm comparison:

ISS 2021 – Scotten W. Jones – Logic Leadership in the PPAC era

So if my numbers are correct if this is true they should be at around or between 65k-75k wafer starts per month in TSMC 7nm alone.

"3nm GAA is Samsung's chance to surpass TSMC, since TSMC will still be using FinFET for their 3nm node."

I agree 3nm GAA is their chance. They probably out all ressources in their and less on their 5nm FF which is more of a 7nm+++. Their 7nm where comparable but on 5nm they fell behind again. Maybe they are similar to TSMC on density at 3nm but much better on power / performance parameters. EUV capacity is still and advantage for TSMC as they are said to have 50% of the EUV scanners installed.

"The real problem is that GlobalFoundries will stop being useful for budget chips (Zen+ and I/O dies are still being made there, but not for long). AMD could solve that by using TSMC's best nodes for the latest stuff and Samsung's previous nodes for budget stuff."

Indeed. They should have finished their 7nm DUV as it wouldn't even need new scanners. I understand they didn't want to invest in 5nm and further for now but should have finished 7nm DUV.

I was also wondering they didn't license 7nm TSMC once they stopped their own as they have cross licensing agreement with TSMC and did similar with Samsung 14nm before. Would have been a massive advantage for AMD now. And option is GFs12nmLP+ from 2020 which seems not to be a 14nm+++ but a new development node with more like 8/10nm like characteristics. Issue is that it needs new design so AMD cannot just use current 12nm designs. Not sure if it is worth then. Otherwhise they could do a Zen+ refresh on it and call it Duron and also do their I/O on that to safe 7nm capacity.

Man, to be a supply chain manager at TSMC.

Quote from: MBV on March 20, 2021, 12:06:58
They still should use Samsung 3nm GAA and further for future products parallel to TSMC. Capacity at TSMC will never be enough for consoles, GPU, CPU, I/O and FPGAs on the newest nodes now that Apple not just orders chips for their phones and Intel and Nvidia also places their top GPUs there.

3nm GAA is Samsung's chance to surpass TSMC, since TSMC will still be using FinFET for their 3nm node.

The real problem is that GlobalFoundries will stop being useful for budget chips (Zen+ and I/O dies are still being made there, but not for long). AMD could solve that by using TSMC's best nodes for the latest stuff and Samsung's previous nodes for budget stuff.

They still should use Samsung 3nm GAA and further for future products parallel to TSMC. Capacity at TSMC will never be enough for consoles, GPU, CPU, I/O and FPGAs on the newest nodes now that Apple not just orders chips for their phones and Intel and Nvidia also places their top GPUs there.

CPUs, FPGAs and Professional GPUs on TSMC and APUs, I/O, Consumer GPUs and consoles on Samsung.