Complex chip shift from Samsung expected to take Apple 12-18 months - Page 2

Quote:

Originally Posted by nht 

 

And there are over $100 billion reasons they can.  

 

The market cap of several of the pure-play foundries are in the single digit billion range.  For example UMC is only $4.7B and is sometimes the rumored takeover target of Qualcomm.  Of course they NEED about another $8B (if I recall correctly) in CAPEX to get to the 20nm process.  TowerJazz is only 181M if all apple wants is to acqui-hire a company that is already in the foundry business (although not in the CPU one they are a ARM licensee) to manage new fabs.  That's a longer road than buying UMC but UMC has a asston of 40nm stuff Apple probably can't use.

You miss the point. Apple keeps that much cash (and other liquid assets) for being able to make investments quickly, that is: being flexible.

If you tell me you want to invest billion in a production plant, that means exactly the opposite. Flexibility is at Apple's heart, that's what they have always been used to!

You must always consider the culture of a company. A company that has always followed, from the very beginning, a "make and buy" approach, and is currently really successful, can't change its colture so easily!

They decided to not be integrated upstream, because that reduces flexibility (see the Nokia case: they were so focused on containing costs and standardizing components that they missed the change in the market). They preferred instead a downstream integration (with the DOS and the Online Store), to be closer to the market and consumers.

Remember that Apple did, in fact, tried to build its own manufacturing plant, in the early years... And that idea was a complete failure. Because they simply weren't used to that! They could have hired hundreds of new engineers, but that would have not helped.

See also what is happening with iCloud: even if it's still more reliable than Mobile Me, it often goes down. Because you may have the best people working, but if you're not specialized in online services (because that is not in your core competencies) you *will* have troubles.

In the past, authors who were supporting the "make" approach, used arguments like: it gives you more margins (but yet, Apple already has huge margins: EBIT margin of 35%) and fits best your needs (but most components are standardized, and Apple works together with its partners to develop them).

And when you have an impressive ROI of 37%, which is ~2.5 times the industry average, and thus a proven competitive advantage, why change?

This doesn't mean, however, that Apple can't invest in its manufacturers. Actually, they should acquire stakes, to gain tighter control (and also help avoiding the issues with breaking contracts). However, they don't need the majority of the company.

PS: I'm a business student, graduating next summer. I may not know a lot about technical details of chips (I know nothing, actually), but I deal with topics like innovation management and supply chain management every day or so...

--EDIT
@Melgross got the point. That's exactly what I was thinking about when I talked of "corporate culture".

Quote:

Originally Posted by melgross 


I can't go with that. It's a very complex business. Just because you think Apple needs to do this doesn't mean that they can do this. Even Intel is finding it difficult to invest in new fabs by themselves. It's thought that the next generation may cost close to $12 billion a pop. You haven't given a cogent reason why they should do this.
There are a number of companies with decades of experience in building and running fabs. Apple has none. Even those companies have problems meeting their standards. TSMC often has problems, and runs behind with every new process technology. Look to the problems Nvidia and ATI have had over the years. Golbal foundries is worse. IBM is pretty good, but doesn't have the production capability. Intel is the best, and can do whatever they need to.
Apple pays suppliers billions every year to upgrade plant and equipment, hire and train workers, and buy equipment for their own use. It's believed that they recently made a $1 billion payment to TSMC. TSMC has been getting better at meeting schedules, and is the largest independent fab out there. It's believed that they could devote one or two fabs for Apple production.
It would be insane for Apple to go this alone.
Even Samsung buys SoC's from other companies for their own devices. Apple's biggest bottlenecks are in displays. So you might as well say that they should build their own plants for that as well. And what about cases? They've had problems with those too.
By not building those by themselves, they have saved billions of dollars. They don't need to deal with plant and infrastructure. They don't have to deal with the workers. They don't have to deal with old equipment and breakdowns. I could go on. This saves them a lot of headaches. Being vertically integrated is a problem. Let others have that problem

Let others have that problem? I think you are condemning Apple to bottlenecks for life. Apple is at maximum production right now. If Apple is really leaving Samsung, then Apple will have to front another manufacturer big big bucks ($1 billion will be a drop in the bucket) to take on the risk of increasing production capacity to Apple's needs. TSMC or any other manufacturer may cherish the thought of having Apple's business but they are also not stupid. Fortunes come and go all the time in the tech world and for a company to tie its future too tightly with Apple would be downright silly.

Samsung is vertically integrated and is in a much better position than Apple for long term health of the company. Samsung will eventually eat Apple's lunch if current trends continue.

Just because you think that Apple can't do this doesn't mean it can't be done (or shouldn't be done).

By the way... I don't recall saying that Apple would have to go it alone but I have said that Apple needs much much more control over chip production. it sounds like you assume I meant that Apple has to go and build its own plant and forget about all the others. Nothing could be further from the truth.

... and, yes, Apple might have to gain more control over its displays as well.

Quote:

Originally Posted by melgross 


Even Samsung buys SoC's from other companies for their own devices. Apple's biggest bottlenecks are in displays. So you might as well say that they should build their own plants for that as well. And what about cases? They've had problems with those too.
By not building those by themselves, they have saved billions of dollars. They don't need to deal with plant and infrastructure. They don't have to deal with the workers. They don't have to deal with old equipment and breakdowns. I could go on. This saves them a lot of headaches. Being vertically integrated is a problem. Let others have that problem

But that has nothing to do with Samsung's manufacturing capacity. Samsung buys Qualcomm's snapdragon SoC's for US LTE only.  Most of Samsung devices use Samsung's own Exynos SoC's.  And let's not forget, Qualcomm contracted Samsung to produce the S4 when TSMC couldn't keep up with Qualcomm's demand. Samsung is now also producing the S4 SoCs.  

Quote:

Originally Posted by tooltalk 

 

But that has nothing to do with Samsung's manufacturing capacity. Samsung buys Qualcomm's snapdragon SoC's for US LTE only.  Most of Samsung devices use Samsung's own Exynos SoC's.  And let's not forget, Qualcomm contracted Samsung to produce the S4 when TSMC couldn't keep up with Qualcomm's demand. Samsung is now also producing the S4 SoCs.  

... and yet Melgross believes that somehow or another that TSMC is going to magically take over from Samsung to produce Apple's chips. Hell, even Samsung is at the top of the capacity that they want to give to Apple.

In other words, these guys want Apple's capacity to go no further than the maximum we're seeing right now. Apple might have 10% left before it hits the ceiling.

Great business plan.

My opinion... Apple is going to have think long and hard about the future of the company... but if Cook really is the numbers guy that everyone says he is then he's already staying awake at night thinking about this.

I see a major amount of cash burning in the very near future.

Quote:

Originally Posted by melgross 



Apple pays suppliers billions every year to upgrade plant and equipment, hire and train workers, and buy equipment for their own use. It's believed that they recently made a $1 billion payment to TSMC. TSMC has been getting better at meeting schedules, and is the largest independent fab out there. It's believed that they could devote one or two fabs for Apple production.
It would be insane for Apple to go this alone.
Even Samsung buys SoC's from other companies for their own devices. Apple's biggest bottlenecks are in displays. So you might as well say that they should build their own plants for that as well. And what about cases? They've had problems with those too.
By not building those by themselves, they have saved billions of dollars. They don't need to deal with plant and infrastructure. They don't have to deal with the workers. They don't have to deal with old equipment and breakdowns. I could go on. This saves them a lot of headaches. Being vertically integrated is a problem. Let others have that problem

When was that? I remember rumors that such a bid was rejected. I can't find the Appleinsider link. Here is one from Macobserver.

Quote:

Originally Posted by jragosta 


I guess you don't understand chip production then. While Intel's current chips are clearly more expensive and more power hungry than ARM chips, that's a function of architecture. If Intel were to act as the foundry for an Ax chip designed by Apple, it would not suffer any power dissipation problems (nor would Intel be able to price it much higher than market price). In fact, since Intel's process technology is the most advanced in the industry, an Ax chip produced by Intel would almost certainly have lower power usage than the same chip produced by Samsung (and, since chip cost is proportional to chip area, Intel's chip might well be less expensive, too. Even if Intel insists in an 80% gross margin (compared to 60% on their existing products), it would still not be more than twice the cost of current chips - probably less.
Now, I agree that Intel is unlikely to fab chips for Apple, but it's not for the reasons you cited.

You have to wonder, though, what the initial cost would be for Intel to begin fabrication of this Arm chip.

Quote:

Originally Posted by nht 

Except that apple has made billion dollar capex in the past to guarantee capacity.
And investing 10% of the cash hoard stuck overseas in a long term strategic move is not unreasonable.
Take UMC...buy UMC with non-USA money, build fabs in Austin using UMC expertise. Essentially you got about $2b of value back into the us untaxed. AND you have the added PR bonus of US jobs creation...and chips are one of those things we can profitably manufacture in the US.

Again, UMC builds chips. Apple "builds ideas: they develop and design hardware and software, but they don't manufacture it.

So whatever argument you take about economic convenience is pointless.

If you want some theory... That is the basic of corporate finance! When considering an investment, you have to look at it under 3 points of views:

1. economic convenience (if the NPV of the investment is > 0 - or other methods)

2. financial sustainability (how to finance the investment - debt or equity? - and the costs of the financing)

3. general impact on the company

Let's say that building chips in-house is economically convenient, as you can get higher margins on a strategic component and you can reduce transaction costs. I take this for granted. Apple shouldn't have big troubles in financing it as well, as they do have a lot of cash (and the "opportunity costs" of using that cash are likely lower than the benefits).

It's in the third point where the investment plan fails. There are hundreds of example of companies that made big investments that were excellent under the first 2 aspects, but proved to be a total failure in the following years. That is because, again, Apple doesn't have any experience in manufacturing. What if the chips were more expensive than what planned? Or, even worse, what if they were faulty (not unlikely, if they aren't experts... again, see what's happened with Mobile Me and what's happening with iCloud, which is still pretty unstable). Even, even worse: what if they weren't able to produce volumes that are big enough? You know: you can't start producing 100 million chips per year from zero... The impact on the statements of income would be bad as well: investing in plants means increasing the operating leverage, and thus making the cost structure more rigid. Eventually, having your own manufacturing plant means that you will have to use it: you are putting a lot of money into a technology, without being able to change it completely in the next years.

Originally Posted by Nathillien 
Are you saying Apple design looks like crap?

No, but if you'd read my post you'd know that.

Quote:

Originally Posted by melgross 

AMD has been a loser company for as far back as I can remember. 

There was a period where the Pentium 4 sucked and the Athlon was supreme.  They were pretty much selling every CPU they could make and it was thought that Intel was on the wane.  

Then came the Core 2/Core 2 Duo.

Bad if your demand craters right after you spent a lot of money investing in fabs yes.  In any case GF is the #2 pure play foundry today.

It's an equally big risk if they are unable to make sufficient iPhones and iPads because their desired foundry has elected to serve another customer.

And I disagree.  If they acquired UMC they would own the #3 pure play foundry in the world behind TSMC and GF with working 28nm fabs and a road map to 14nm FinFET.

They could do that and STILL partner with Samsung and TSMC in the near term for 10% of their money hoard.  Somewhat less if they sold off the 40nm capacity to someone since they likely wouldn't want it.

Going the IDM route is a viable one for Apple and they certainly have more than enough money and technical expertise to do so if they choose.  I bet they do so within the next few years.

Quote:

Originally Posted by sr2012 

[SNIP]

The fact is that 32nm single core Atom Saltwell isn't bad as a smartphone processor and on par with the A9. The late 2013 22nm quad core Bay Trails using Silvermont should be pretty impressive and the Clover Trail Saltwells are probably on par with A15s despite being a process node behind.  The older 18W TDP 45nm D525 was slower and more power hungry than the ARM Exynos 5 (around 4W TDP) in the Phoronix benchmarks but the 32nm Z2760 has a TDP around 1.7W should be on par with the Exynos in terms of performance per watt even if slower overall.  

Intel's TDP has been plummeting over the last few years.  Low power Haswells will have 10W TDP...nearly half that of the D525 Atoms.  Current gen low power Core i5s and i7s are in the 17W TDP range.

Counting Intel out today is as stupid as counting them out when they looked really bad against AMD with the P4/iTainic debacle.  Intel has zero need to make ARM chips given the massive improvements they've made the last few years in power performance.  The 2016 mobile CPU landscape (both smartphone and tablet) may look as different from the 2012 landscape as the 2006 desktop CPU landscape looked from the 2002 landscape when the AMD's Athlon XP was kicking around Intel's Northwood P4 processor like a sad little puppy.

If Intel's roadmap continues unabated I could potentially see the 2016 iPad make the hop to Atom just like the Mac moved from the PPC despite Apple's investments.  The 2014 14nm Airmonts indicate to me that Intel is moving to tick/tock for Atom as well Core and putting Atom on the same process node.  

Quote:

Originally Posted by sr2012 

@nht OK, I could be wrong. But something is fishy about Intel and Medfield. Benches are one thing, design wins another, adoption by the major players even another.
It looks promising, but in the long run (5-10 years), I am still not sure Intel can do such a big turnaround.
http://seekingalpha.com/article/353351-i-d-still-take-arm-holdings-by-a-nose

He says this:

"Intel is using the billions it spends on proprietary processing technology to wage a 'die shrink' war.

What it lacks in low power system on chip design expertise, it makes up for in being able to make smaller chips that those who rely on foundries such as Taiwan Semiconductor (TSM) and United Microelectronics Corporation (UMC)."

And then doesn't realize that Medfield is 32nm while many 2012 Cortex A9s it competes with are 28nm as are the Cortex A15s.  Atom has traditionally been a process node or two behind Intel's best. That wont change until 2014.  If Intel had given up some 2012 Ivy Bridge production capacity for 22nm Atom production then this analyst might have a point.  But it didn't so he doesn't.

Intel spends billions on process technology because that's part of being competitive in the chip business.  Especially in the far more profitable desktop and server CPU business.  ARM is the new AMD...it's keeping Intel prices in check and that's a good thing.  It's beating Intel in the power department leading Intel to invest heavily in improving the efficiencies of their Core design and that's a good thing if you like using Mac laptops in addition to your iPad.

Wouldn't it be amusing if after finally completing the shift, Samsung were to launch a hostile takeover of TSMC.

Quote:

Originally Posted by sr2012 

Yes Intel is single-core posting good benches but it does seem dual and quad ARM A9 might not be optimised well.

The thing about multi-core is that it benches better than real world performance in general.  Why?  Because parallelizing your code is hard to do right.  You can have worker threads to do whatever but most programs are still fairly linear beyond moving some time consuming tasks off the main GPU thread.  There's simply not that much trivially/embarrasingly parallelizable problems that are easily solved to move to multi-core.  Kinda like GPU computing.  I would be 99% of real world GPU computing is via PhysX and not CUDA or OpenCL.

So the single core design of the current Medfield is not as big a hindrance as it would seem even with quad-core ARMs appearing.