Remember also that everybody apparently wants to get into the ARM server business and none of them are aiming for $11bn revenue and yet they still intend to make a profit at it. I know what you said was just a throwaway remark in your article based on perceived wisdom and that which may have been true initially and historically but it did reveal a poor grasp of current server market costs/margins that I just had to bring up because it will have relevance for you going forward in attempting to cost the ARMy when they try it.
'I still think op margin would drop from ~50 to mid/high 30's. '
You might 'think' it but you have not shown any data or analysis to prove it, in fact all the evidence from various sources points the other way. Intel has over 90% of the lucrative server market and you 'think' that is only worth 30% operating margin ?! What do you think all the others survive on ? Seriously this is supposed to be your 'job' now and I have shown that you are basically just guessing here and guessing quite wrong.
Remember though your article explicitly said it was the synergistic benefit of Core PC chips that allowed DCG such high margins, Atom or Phi were not part of your original statement on which we started this debate. Atom and its server derivatives would be dropped in my version of DCG anyway as it has no performance/power advantage over LV Xeons and just adds cost for little or no server benefit.
Not any more now it has grown to be such a large self-sustaining business. It might even be hindered by it as they are stuck with whatever Core design they are given which is obviously client orientated. When it was a small business and Xeon was just a Core with extra cache and RAS and volumes and asp was lower, sure it allowed DCG to survive undercutting everybody else and establishing market share just like Mobile is doing now starting out. Now though because Intel is the Big Dog and charges high prices like the Riscs did (and still do) and has such high market share and revenue it could survive on its own.
Not at basic circuit level where cache coherency has to be worked out and electrical voltage planes and a whole host of other low level integration issues have to be worked out. The various operating segments might make different skus out of them which is their cost but having a cpu talking to a gpu successfully on the same die would already be done for them. Even if you choose to ignore all this, which would be foolish, you still have to share the Shared grouping between Core, Atom, Phi and Gpus. You are not going to get Core development up to be $1bn no matter how hard you wish for it to be so ;-).
Yes and it says 'Shared Processor and Gfx' which obviously includes integration between the two which would go in the new Server only Intel. I would not put Core as any more than $500m annually especially as Atom architecture is also included in that grouping as well as gpu and circuit integration between all of them. Another $500m for DCG's specific R&D gives you $1bn total annual server processor development allied to a reduced $1.5bn process R&D gives my new total $2.5bn R&D figure but quite frankly the process R&D could go too as you could just make the chips at TSMC now the margins are so great.
You should remove ALL as combined cpu/gpu dies would be totally out of the equation. Are you not understanding this basic point ?
I would assume GMs closer to 85% (~$400 asp for ~$30 dies) and R&D around $2.5bn as I would not add anything from shared as it would all go in the new server only Intel and the process development schedule could be relaxed. SG&A, I would put as $1.5bn so ...
Revs: $11 billion
R&D: $2.5 billion
SG&A: $1.5 billion
Gross margin $: 9.35 billion
operating margin = $9.35B - $2.5B - $1.5B = $5.35 billion = 49% of revenues.
So I actually expect a slightly higher operating margin which you would expect with concentrating on just a premium high asp product.
'So, let's say half of Intel's "shared" processor + GPU R&D is half GPU and half CPU to play it safe. This implies that the processor R&D per year is in excess of $1 billion dollars.'
Servers do not require gpus so you can un_add that half cpu of shared. Again you are adding development related to Core client processors to Xeon.
'If it took $200M-$300M to support the development of one 32nm CPU + accompanying chipset per year, then how much do you think it would cost to support: '
32nm Poulson was a brand-new architecture and chipset compared to 65nm Tukwila so your attempts to characterize it as 'accompanying is yet more sloppy falsehood. I would even characterize at as a lie if I knew for certain you were doing it intentionally. Also the Itanium pipeline has always been more than 2 deep, Kittson was no doubt being worked on as well as it was mentioned around that time so they must have had an idea what it was, i.e. originally a 22nm shrink. The Itanium model is the one most relevant to the idea of a server only Intel as there would be no need for SoC, gpu, Atom extraneous development with which you are trying to justify your erroneous $1bn figure which I might add was originally to answer just how much it would cost to add Core Architecture to DCG whereas now you are trying to mislead by adding everything else that Intel does to DCG when most is not relevant. Also a lot of your 3 to 6 is already included in the R&D cost of DCG, are you adding it twice now to get to your 1bn figure ? Just how disingenuous do you actually want to be here to defend your erroneous number ? Your point 1 is the only thing your $1bn figure should be addressing and clearly it does not match.
'$1 billion is not unreasonable at all.
Of course it is if we are just talking about the *additional* contribution of Core development to what is already spent on DCG. Remember the Tukwila/Poulson money also included designing the core architecture as well as the chipset which DCG currently does not have to do. Your fundamental error of analysis was not dividing what it takes to develop a cpu/soc (~$200-300m) by the number of years it takes (3-4) so you got a number that was 3-4 times as high.
Thanks for your considered opinion. I was just trying to get a a handle on what you as an experienced long thought was happening at a very low level and I thought I would present the cold financials to you and see what you made of them. I agree with you if GMs can be stabilized and then increased and all the expected new business materializes as planned the company and stock should have a bright future. Getting into quad-cores and tablets is also good TAM expansion and the prices look very competitive for the product BoM that's being offered although it would be informative to potential buyers if the actual processor chipset info was given to the public, apparently it's an A7 quad-core Snapdragon 400 in the top model. Good luck to you and all IFON shareholders and who knows I might become one one day if I am convinced the GM issue is stabilized for good ;-).
'33% of the smartphone market seems absolutely impossible'
Remember this is a 10-year forecast. If Intel can maintain a 30-40% SoC performance/price/power advantage going forward and taking into account that a) Samsung/Apple will continue to do their own thing b) phones sell on design and fashion as much as on SoC abilities c) some phone OEMs will just continue to ignore Intel regardless then it does not seem an outlandish forecast based on taking market from the ground up the hard way by virtue of being better and cheaper. Now if Intel did not have those 3 impediments I mentioned it would have been more like 50-60% IMO.
Now maintaining a 30-40% SoC performance/price/power advantage will depend heavily on Intel solving the industry lithography show-stoppers below 10nm satisfactorily and as well as anyone else as I am quite sure x86 is the superior ISA on an equivalent or better process. This is the biggest long term issue facing the processor industry right now as 193nm UV lithography has seemed to be where the industry has halted its progress for some years now.
TSMC 20nm wafer production delayed temporarily, but will not affect shipments
Josephine Lien, Taipei; Steve Shen, DIGITIMES [Tuesday 11 March 2014]
Taiwan Semiconductor Manufacturing Company (TSMC) has confirmed that its 20nm wafer starts were delayed recently due to a specification issue of materials used in its CMP (chemical mechanical planarization) process. The issue has been resolved, and no shipments will be delayed, said the wafer foundry house.
TSMC began volume production of its 20nm process node at the beginning of 2014, focusing on Apple's A8 processors, according to industry sources.
The yield rate of the 20nm process has been ramped up significantly since then, reaching a level ahead of schedule, the sources noted.
While the production of 20nm wafers was delayed temporarily, no production lines have been shut down partially, and the CMP materials issue has been solved since TSMC has been maintaining a multi-source policy for the purchase of needed materials, TSMC said.
TSMC will continue to ramp its 20nm capacity quarter on quarter and expects sales of 20nm products to account for 20% of its total revenues in the fourth quarter of 2014.
'If I'm recollecting correctly, IFON released 4 smartphones since May of 2013, two of them toward (or in) the 4th quarter (just reported). '
and yet margins still fell from the 3rd to 4th quarter ...
Higher revenue and income than today which was also growing year on year. Basically an end game where Intel controls 50% of the tablet market and 33% of the smartphone market as well as 90-95% of the PC/Server market (sorry AMD/IBMers). The opposition will be left behind by 10nm Core and Atom.
The fundamental point you have to grasp is this, DCG's high profitability is because they are selling billions of dollars of $20-40 dies for $200-2000. They could even go fabless on a business like this and shunt $20-40 TSMC's way too and it still would be viable. The fact there is so much revenue now of such a high margin means the business could be self-sustaining now without Core. It wasn't always like this at the beginning when revenue and margins were much lower but now it is.
['We bow down to all you amateur and professional financial analyst BS'ers'
Very unprofessional. I was enjoying our debate which was free of name calling and personal attacks, but then you had to punctuate it with this 'gem'. ]
It's a gem because it is true. In the fantasy analyst world you and guys like you inhabit reality is what you say it is not what it really is.