Why Lunar Lake changes (almost) everything
- Thread starter TICin852
- Start date
Which is TSMC, for all the reasons I said... ;-)
There is also Samsung so even if Intel decided to shut down the fabs, there would not be a monopoly. Like you said, keeping up on advanced nodes requires money and a talented and hard working engineering workforce, which Samsung has access to due to their cross subsidies and their geographical location.
That's true, but Samsung's leading-edge processes have been uncompetitive in recent years, including on cost/yield -- they're great at making exciting technology announcements (first to HNS!!!) but not so good at either getting competitive PPA out of them (their 3nm HNS is less dense and higher power than TSMC N3 FinFET) or getting yield up to competitive levels. This has been going on for several generations now, it's not just a one-off.
So yes, TSMC aren't a monopoly -- unless you want the smallest lowest-power highest-yield chips, in which case there's currently no effective competition... ;-)
I understand your points:
But can't Intel just do some financial engineering to move money around or even take some of the profits from some BUs and transfer that money to other BUs? They are still allowed to have asset transfers between BUs. They could call it a prepay if they even wanted to. Or an interest-free loan. Or something else to get some money to the IF BU.
I do understand your point however: Intel has to pay for the process node somehow - and if their margins are only the same as AMD, Nvidia, etc. they won't be able to compete against TSMC - Intel still has to pay to develop a node AND products. The vertical integration improves their gross margin compared to fabless: but by how much? - and is this amount enough to pay for the fabs and nodes?
Here's a few questions I have though: QCOM seems set on taking laptops, and if Intel is able to at least make a decent pivot to AI (doubtful that will happen given their past failures - but it seems they are at least trying right now) they compete with NVDA - these are the two customers that seemed most likely to be using IF in the future when it was announced. Should we consider IF dead then? Are these competitors willing to at least (in some way) pay for Intel's node development which would indirectly help their competitor, Intel? How much do they value not having a TSMC monopoly?
Thank you for sharing your experience with Fujitsu. It seems like you see many parallels between Fujitsu and Intel. Based on what you have seen, what would you advise Intel to do differently? Is it even possible to avoid Fujitsu’s fate?
hist78
Well-known member
My observation is that Intel operates with wrong business model, high cost, wrong product mix, and/or bad strategies. Take a look on the table I created below. TSMC maintains a decent net profit margin while most TSMC's major customers are doing well. On the other hand Intel's (and AMD's) major customers, such as Lenovo, HP Inc., DELL, Asus, and Acer are always making meager net profit. In order to compete against TSMC and Samsung, Intel Foundry service may need to cut the price and cause Intel's net profit margin to go down further.
Last edited:
tooLongInEDA
Moderator
This is an interesting insight.
Intel in part became hugely successful through its ability to hoover up almost all the profits from the PC industry. But low margin customers doesn't leave any room to squeeze further.
On the other hand, TSMC's customers are mainly running high margin businesses. And without the need for TSMC to squeeze their margins down as happened in the PC business. You can't, ultimately, be successful if your customers are not.
Thinking about it this way, IFS is an opportunity for Intel to acquire some of those higher margin, higher growth customers.
hist78
Well-known member
"Thinking about it this way, IFS is an opportunity for Intel to acquire some of those higher margin, higher growth customers."
Yes, That's one of the possibilities. But many of them are Intel product division's competitors. They are not in any charitable mood to help Intel to improve its profit margin and financial standing. Actually they may use TSMC or Samsung as the leverage to cut Intel's price to the bottom.
As I said, they need to find customers who are willing to pay a premium for things like US manufacture and security of supply -- and they'll presumably be able to get a chunk of funding from the government for this, which is justified given the current uncertainty about China's likely future with Taiwan.
But the basic problem is that "IDMv2" (or "IDM-lite", or whatever you want to call it) basically doesn't work -- an IDM has the advantage of everything being in-house and customized to the product division needs, and this works so long as you have a big enough volume of high-margin custom products to fill the fabs and a competitive process, it's what Intel did successfully for many years. But it stops working if your process and/or products fall behind due to either a bad process (like Intel 10nm) or bad products, and then panic sets it -- how can we fix this?
A foundry like TSMC is the complete opposite, it supports a huge number of diverse customers and has the support and IP ecosystem set up to do this -- and there's no way Intel Foundry can compete with this starting from where they are, either on price or support or investment.
What Fujitsu did -- and what Intel are trying to do -- falls in between the two camps. Once split up and opened to the outside world they can no longer use the IDM advantage of tightly-coupled product and process/IP customised for it, including difficult design flow and tools (like Intel used to do) to gain an advantage, but they also can't compete with a much bigger pure-play foundry on their playing field. There's also the problem that some of the big customers they want/need to get are competitors to Intel product division, which is not exactly going to help either.
In my view Intel had two choices -- either double-up as an IDM and try and catch up process-wise and use the advantages of being tightly integrated (old-style Intel), or completely spin off the foundry and let it sink or swim and leave the product division completely free to choose which foundry they want to use, with no pressure to stay with Intel Foundry -- and it costs a lot of time/effort/money to support two foundries at the same time, which is why most customers choose one and stick with it, at least as far as bleeding-edge chips which may need a lot of custom IP are concerned.
It may be that the second is what Intel are actually going to do, but it doesn't look like that given the way that the product division is still pushing the claimed advantages of using Intel Foundry -- and it's not all untrue, their push to have backside power first is the right thing to do for products like x86 CPUs. But it seems unlikely in the extreme that Intel Foundry can change and expand to be a real competitor to TSMC, though with critical customer and government support (if they can get this) they may well survive and even succeed as a smaller niche US-centric foundry, perhaps targeted at CPU/HPC products.
But against this TSMC are also building fabs in the US which have all the TSMC advantages, and even if the wafers cost more than ones from TSMC Taiwan they're still likely to be cheaper than ones from Intel Foundry, and they have the massive PSMC IP ecosystem behind them -- so they're still likely to get more of the US-centric/secure supply business.
So it's possible that Intel can avoid what happened to Fujitsu, but the foundry will have to succeed based entirely on its own merits with little support from Intel product division, who are likely to use TSMC for most products for the same reason everyone else does -- and these products will also have to win against competitors like AMD and Qualcomm using the same foundry, so again purely on their own merits. Given that AMD adopted chiplets (the future) far earlier than Intel and Qualcomm (and others) aren't carrying a lot of heavy x86 baggage, that's also not going to be easy for Intel in the future.
If Intel hadn't had such a major screw-up with 10nm they wouldn't be in this mess, but that's water under the bridge. They might have been better staying as an IDM and using the advantages of tight integration, if the foundry could catch up with TSMC on process -- which may well happen, at least in the short term for CPU-type products -- but then that's water under the bridge too now they've panicked and split the two divisions apart.
In the end it comes down to bad/cowardly management -- too scared to go for broke as an IDM, and not brave enough to completely split off the foundry and let it sink or swim. Which I suspect they'll have to do in the future, both to help the product division succeed (not tied to IF) and the foundry (not tied to Intel CPUs for competitors).
The problem with moving money around like that -- even if it's legal and isn't challenged by shareholders -- is that is takes profit away from the product group which they need to use to invest in and develop new products, so why should the product group agree to this? Their best option *as a product group* is not to subsidise IF but to choose whichever the best foundry is for their products, whether this is IF or TSMC.
The real advantage of vertical integration like Intel had is that it allows customisation of both process and chip design to get better products (e.g. higher clock speed), which then allows you to either sell more products or at a higher price, especially if you have a stranglehold on a large and lucrative sector, so it doesn't even matter of the wafers are more expensive (which they were). This worked well for Intel for a long time with process and IP specifically targeted at their CPU designs -- until they screwed up 10nm and it all went horribly wrong... :-(
I appreciate the detailed and thoughtful answer. Intel management justifies going into the foundry business to amortize the growing costs of development so the IDM model might have broken down anyway. Instead of going into foundry, I do wonder if Intel could have secured a sweetheart deal with external foundries if they had agreed to shut down the fabs and disperse the organization to kill off competition.
Xeon is high perf, Lunar is low power. Allows Intel to focus the 3 process on one set of metrics. It looks like 20A/18A is where it all comes home and will need to be more versatile.
It is a phantom advantage. Much better for foundry to hear the needs of more customers - and have more customer volume - and for the design side to have all choices available. "vertical integration" is really "three-legged race".
And they have different capital structures, timelines, staffing, and will appeal to different investors. The private capital deals seem halfway to spinning off the foundries, and will likely strengthen the drive to do just that. Those companies are most likely looking for the day their investment cashes out, not simply a steady dividend.
.. and were paranoid secretive and unable to even admit to each other how bad it was, for years. The new model inherently prevents that.
siliconbruh999
Well-known member
Review of Xeon 6 Sierra forest on Intel 3
Last edited:
Good points. At this Goldman Sachs conference a few months ago Intel (I think the CFO Zinsner) told us that the ASPs for 18A will be 3X/wafer - here it is.
I also heard Zinsner (or it could have been Gelsinger) repeat that in their recent conference call - it could have been either Deutsche Bank (Pat) or Citi (David).
Just going from memory there.
Foundry:
- Our plan of record is to intercept High-NA at Intel 14A. I'll also remind you that if for any reason, High-NA is not production worthy at that point, we can still move forward
- with Intel 14A, it's fully backwards compatible with just an EUV process
- We've talked about getting the 1.0 PDK for 18A out this quarter. We've got products in fabs that will be ramping middle of next year for release in the second half of next
- year with Clearwater and Panther Lake because really what we're counting on is a mix shift of -- in our wafer capacity from uneconomical Intel 7 to very economical Intel
- 18A and the ability to pull tiles back in if you look at the move from Intel 7 to Intel 18A, the ASP per wafer goes up almost 3 times
- The cost per wafer doesn't change that significantly. And it really kind of illustrates how uneconomic the Intel 7 process is without EUV, with all that multi-patterning>>
Tom's Hardware is saying that Intel could get up to $100 billion in tax breaks (?) on top of the grants and low-interest loans.
The USA wants Intel to sign up the 'whales' (NVIDIA, Qualcomm, etc.) so Boeing, Northrop Grumman and more can
get cutting-edge capacity for the first time ever. And Jensen said that he'd like that too. There can be a lot of winners.
https://wccftech.com/nvidia-teases-...e-third-foundry-partner-besides-tsmc-samsung/
<<Intel is one of the biggest recipients of Biden’s push to recharge the American semiconductor industry, with $8.5 billion
earmarked for Team Blue, plus another $11 billion in low-interest loans. It also received a 25% tax credit for up to $100 billion.
This massive amount doesn’t come without strings attached, though. >>
https://www.tomshardware.com/tech-i...rmation-before-disbursing-billions-of-dollars
https://www.intel.com/content/www/u...ategic-defense-industrial-base-customers.html
Reuters says $25 billion in tax breaks - Tom's Hardware (in the link above) says $100 billion and that just seems too high to me.
https://www.reuters.com/technology/...nding-spree-across-four-us-states-2024-03-20/
Last edited:
Jensen doesn't want a TSMC monopoly - no one does:
https://wccftech.com/nvidia-teases-...e-third-foundry-partner-besides-tsmc-samsung/
Intel sold a 49% stake on those factories. They are keeping 51% of the capital and control of the fabs. So it is not as bad as you are talking about. Intel will then use this money from the sale to complete their new fabs in the US. The plan makes perfect sense in my opinion.