You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
NVIDIA CEO Jensen Huang shared key details for his firm's future yesterday at the Goldman Sachs Communacopia + Technology conference. Huang's comments appeared to be just the catalyst that NVIDIA's shares needed, as the stock gained 10% yesterday and opened higher today as well. During the talk, he commented that even if the broader industry does not adopt artificial intelligence, the demand for NVIDIA's products will still be sustainable simply because of the physics surrounding semiconductor fabrication.
He added that NVIDIA has developed its semiconductor design capabilities so that even if its manufacturing partner TSMC were to face disruption, NVIDIA could continue its GPU supply by switching partners.
Nvidia can and should diversify as should Apple, Qualcomm and AMD. One could argue the others like Google, AWS and Microsoft may not have enough volume to justify but certainly enough money.
If Intels delivers on 18A even marginally and 14A it is a no brainer to do your main volume at TSMC and enough at Intel to keep them alive.
Just like buying AMD and Intel in the 90s and the last decade.
Dual source is good business and smart business continuity.
After all a super typhoon or big earthquake can ruin your business for a year plus
Nvidia can and should diversify as should Apple, Qualcomm and AMD. One could argue the others like Google, AWS and Microsoft may not have enough volume to justify but certainly enough money.
If Intels delivers on 18A even marginally and 14A it is a no brainer to do your main volume at TSMC and enough at Intel to keep them alive.
Just like buying AMD and Intel in the 90s and the last decade.
Dual source is good business and smart business continuity.
After all a super typhoon or big earthquake can ruin your business for a year plus
The irony is that it is Intel themselves who perhaps did more than anyone else to kill off second sourcing in the semi business when they decided to stop dual sourcing with AMD from the 386 (about 40 years ago).
But dual sourcing has to be a far more difficult process these days. You can't just ship the masks to another foundry. Nor can you replicate a modern chip (too complex, too much software).
Is there an argument to be made that dual sourcing (e.g. sites in different continents) with a single supplier might be a better practical option than dual sourcing across different suppliers ? In the 1980s, defence contracts often mandated at least dual sources - specfically meaning independent companies. But is that what we really need today ?
A lot of the discussion around Intel seems to start out from rather fuzzy - and it often seems - sentiment/emotion-driven goals. I'm not yet convinced that the US government and CHIPS Act have got to the core of the real needs here.
Nvidia can and should diversify as should Apple, Qualcomm and AMD. One could argue the others like Google, AWS and Microsoft may not have enough volume to justify but certainly enough money.
If Intels delivers on 18A even marginally and 14A it is a no brainer to do your main volume at TSMC and enough at Intel to keep them alive.
Just like buying AMD and Intel in the 90s and the last decade.
Dual source is good business and smart business continuity.
After all a super typhoon or big earthquake can ruin your business for a year plus
Nvidia is doing chiplets now so dual source is much easier than full chip, one of the many chiplet advantages. Apple is still full chip and I do not see that changing for mobile SoCs anytime soon. Everyone one else is chiplets (AMD, Microsoft, Google, AWS, etc...) so second sourcing is a real opportunity. Packaging will then be much more important so Intel Foundry could get more business since TSMC will not package other foundry die, yet.
Nvidia and the others can throw a chiplet to Intel and use their packaging and Intel can claim whale customers. Unfortunately the financials behind it will disappoint investors due to the lower margins etc... but it is a start in the right direction.
Nvidia is doing chiplets now so dual source is much easier than full chip, one of the many chiplet advantages. Apple is still full chip and I do not see that changing for mobile SoCs anytime soon. Everyone one else is chiplets (AMD, Microsoft, Google, AWS, etc...) so second sourcing is a real opportunity. Packaging will then be much more important so Intel Foundry could get more business since TSMC will not package other foundry die, yet.
Nvidia and the others can throw a chiplet to Intel and use their packaging and Intel can claim whale customers. Unfortunately the financials behind it will disappoint investors due to the lower margins etc... but it is a start in the right direction.
The advanced packaging business can become really crucial and really difficult going forward. We are already seeing NVDA delaying their blackwell chips due to challenges related to packaging. In this particular case, whom to blame is a seperate issue, but it certainly highlights the importance and difficulty of advanced packaging.
I can totally see advanced packaging becoming a high margin business, simply because 1) very few companies can do it well, and 2) it becomes the only practical way to scale up, as we are only seeing 15%-ish PPA improvement with each new process node (N5->N3, etc).
The density gains for N2,A16 are under 20%
Intel 18A is 30% over Intel 3 which is 10% over Intel 4 Intel 3 has lower baseline than TSMC N3 so there is that as well
14A is iirc 20% density increase anyway density is going to be a lot lower from the pattern Advance Packing/Chipset will be the way forward to cram more Transistors
The density gains for N2,A16 are under 20%
Intel 18A is 30% over Intel 3 which is 10% over Intel 4 Intel 3 has lower baseline than TSMC N3 so there is that as well
14A is iirc 20% density increase anyway density is going to be a lot lower from the pattern Advance Packing/Chipset will be the way forward to cram more Transistors
The TSMC OIP Symposium is next week, we should get some update numbers then. According to TSMC, N2 tape-outs will rival N3. I did not see that coming. I thought Intel 18A or even Samsung 3nm would get a share. Hopefully they will get some whales in the angstrom era?
The irony is that it is Intel themselves who perhaps did more than anyone else to kill off second sourcing in the semi business when they decided to stop dual sourcing with AMD from the 386 (about 40 years ago).
But dual sourcing has to be a far more difficult process these days. You can't just ship the masks to another foundry. Nor can you replicate a modern chip (too complex, too much software).
When I think of dual source, that means Dell, HP, IBM ( when they made PC) and many others wanted more than Intel.
Intel wasn’t the most ethical and of course hated how AMD woulld out innovate them again and again. That again shows the need for more than one company as it drives competition.
The TSMC OIP Symposium is next week, we should get some update numbers then. According to TSMC, N2 tape-outs will rival N3. I did not see that coming. I thought Intel 18A or even Samsung 3nm would get a share. Hopefully they will get some whales in the angstrom era?
wccftech.com
