In regards to Apple using TSMC, or Samsung or any other foundry for chip production, it's not the same as waiting for Intel to design, produce and release their chips.
Actually, in substance, it is. Again, you're conflating the design and production steps. The delays in getting certain chip designs from Intel hasn't been due to delays in the design process. It's due to delays in the ramping of the process technology on which those chips were designed for. You don't seem to be understanding that this is an issue of fabrication, not design.
Sure, I suppose there may be a complaint that certain new design features aren't implemented on any Intel designs available on manufacturing processes then shipping in current volumes, but most of those features are either (a) continuing integration into the CPU of additional memory (more cache layers) and migration of northbridge functionality (like PCI-E and memory controllers) and (b) secondary functionality (like more GPU units and video transcoding hardware). The absence of having those features on the CPU represent, at worst (y) a minor decrease in real world memory and bus bandwidth and latency and (z) features that can (as they previously were) unloaded to secondary chips, including commodity and custom designed ASICs, FPGAs, or more generally programmable chips like modern discrete GPUs or special-purpose SOCs (e.g. Apple's T1).
As between one generation and the next, the former has relatively little impact on Apple's ability to introduce new form factors or product. And the latter, which are simpler, easier, and generally not manufactured on bleeding edge process technology, just might have something to do with Apple's interest in expanding its design capability and expertise related to those particular features (e.g. in-house GPU designs instead of Imagination) and chip architectures suited to the purpose (e.g. ARM).
To put it as gently as I can, you seem to be out of your depth. It appears from your comments that you have no more than a superficial understanding of either the technology or business considerations involved in the design, to say nothing of the development and operation of new manufacturing process technology, for high performance logic (as distinguished from, e.g. memory as discussed below). That is not and is not intended to be an affront to your person, intellect, or ability. Please don’t take it that way. It's just (very specific) subject matter expertise that the vast majority of people have little exposure to or knowledge of, not least because most people have neither the need nor opportunity.
Apple, in designing their own silicone, has the ability to work with multiple sources for production, and can dictate when they need deliverable product. They don't have that ability with Intel.
This is sort of what I’m talking about. I’m pretty sure Apple doesn’t design (or manufacture) much silicone. And I’m damn sure Intel doesn’t. Most of the
silicone accessories that you find for Apple products are commodity sleeves and the like designed and manufactured by third-parties (
example).
Silicon, on the other hand…
Or they can certainly say they want product on X date, but when Intel fails to hit that date, Apple doesn't really have any way to recover from the failure.
Again, that’s a process technology and manufacturing issue. And it’s not one unique to Intel. What is unique to Intel, and it has been an issue (most recently and notably, with mobile Skylake parts), is getting the new features that are incorporated into Intel’s new designs, whose launch also depends on the launch of a new manufacturing process, when existing available designs don’t contain those features and launch of the new process node is delayed. Here’s the thing: even if Apple has or could develop designs incorporating those features, it can’t get them on that process technology either. Nobody can. Intel remains ahead of the industry on process technology. So if Intel can’t do it then nobody can.
That matters. At any given time, there isn’t much, if anything, in Intel’s available high-end designs, in terms of features and integration, that Apple
doesn’t want. It’s invariably about something Apple does want (like newer or better hardware implemented video transcoding routines, security features, or GPU functions implemented in hardware). All of that stuff requires transistors and wiring. If that amounts to significant die space (e.g. more GPU units), then it also requires (a) designing in those features at the expense of general computation units or other hardware dedicated to existing features or (b) targeting and using new process technology. A new process node gets you ~30% smaller logic, though anymore the density gains for the entire design are somewhat lower than that would imply (roughly double), depending on the ratio of logic to SRAM, which (now) only scales at about 10% per node. The opportunity to do further design optimizations and layout on an existing node, combined with possible yield improvements from process optimizations, might get the necessary space to do a new design or tweak an existing design to get the new features you want without giving up anything else before a new node is available.
If Apple’s CPU performance and feature wish list is so extensive that it depends on a new node, Apple is stuck waiting, regardless of whomever's design it is using. Just because it can be designed doesn’t mean it can be produced. And if we assume (as I believe we can) that substantially all of the functionality and performance of existing designs is wanted, then the only scenario we’re talking about is the prospect of planning and having ready a design that targets the current node, either on an existing or perhaps optimized process. So far as that goes, Apple only reaps a benefit from using its own designs, over Intel’s, to the extent it could and would plan and develop implementation of those features into designs targeted at available process technology, where Intel, for whatever, has not done so (and therefore may be delayed in doing so pending or until its planned introduction of a new process node).
So, as I say, that’s a problem largely driven by process technology. And on that front, Intel has recalibrated its manufacturing process roadmap to reflect the increasing difficulty and time involved in perfecting and ramping up production for future process nodes. The result is that there will now be three planned manufacturing processes developed (the launch process plus two enhancements) over the life of a given process node (e.g. 10nm, 10nm+, and 10nm++). That will make it easier to plan for and get timely roll-out of planned and requested design features, to the extent they can be accommodated at a given node (see above), even if there is delay in ramping up the next node. This will address or help mitigate launch timing problems like Apple had with products built around Skylake mobile parts.
And to the extent it doesn't (and for other completely unrelated reasons that are particular and peculiar to Apple), Apple can address them more easily, quickly, and cheaply by designing custom ASICs or comparatively simple discrete co-processors, rather than dealing with the complexity, risk, and expense of rolling those features into already complex CPU/SOC parts that have longer lead times and therefore are designed to target planned (future) process technologies.
And the thing about designing and producing chips for laptop or desktop use, the numbers needed are going to be such small volume, compared to iPhone / iPad / Watch / ATV needs that it's not likely to present an issue for any manufacturer.
That’s true, although Apple is already such a large part of TSMC’s portfolio that I’m not sure they’re chomping at the bit to commit even relatively small incremental business. A jump from 20% to 25% of TSMC’s business would be enormous. Even a jump from 20-22% would be material.
Meanwhile, this is an excellent point against Apple wanting anything to do with getting into the fab business itself, whether as an equity investor or, even worse, an outright owner and operator.
Apple's recent acquisition of Toshiba's chip unit is likely a step in the direction of their owning their own fab, or at least one they can use more exclusively to avoid roadblocks in bringing product to market.
Not really, on either point. Regarding “a step in the direction of owning their own fab,” Apple’s ownership interest is so attenuated that what you’ve characterized as a step can only be described as “awfully small.” You should look at the publicly available information about the deal terms.
Tentatively (the deal may fall apart because it looks like the parties don’t agree on what the deal terms actually are and for other related reasons), here are the known terms. Apple is one of
six passive investors, of which the four based in the U.S. (including Apple) will receive
no common stock or voting rights in Pangea, the Holdco created by Bain Capital to acquire all of Toshiba Memory Corporation’s common shares. Japanese investor Hoya and seller Toshiba Corp, will own 9.9% and 40.2% respectively, an aggregate 50.1% of common shares and associated voting rights, giving them control of Pangea. Pangea will continue to be operating by existing TMC management, with input from Bain Capital, lead investor for the acquiring consortium (which includes Hoya). The remaining common shares will be split between Bain and SK Hynix, the latter of which will also obtain convertible bonds redeemable for as much as 15% of the common stock. In short: Apple is a minority passive investor, who will receive no present or convertible control or common equity rights, in a consortia that itself will be a minority interest, as measured by both common equity and control rights. Plus the whole thing looks like it might fall apart, either because the consortium itself falls apart or because Western Digital, one of TMC’s existing JV partners, is pursuing legal action to prevent consummation of the deal.
Setting all that aside, TMC’s business has nothing to do with “bringing product to market” for Apple, where the product we’re talking about in this context is Apple-designed ARM logic. That’s because TMC manufactures NAND memory. So it does absolutely nothing for Apple with respect to any concerns Apple may have about its supply chain for CPUs, whether Apple’s own ARM designs or otherwise.
For the same reason, it’s also far from suggesting any interest on Apple’s part in developing the expertise necessary to own, operate, or develop new manufacturing processes for fabricating its own logic designs. Fabricating logic is more complicated and difficult than fabricating memory, for a host of reasons. On the one hand, it’s a natural place to start in the business. On the other hand, it’s a massively capital intensive business, something to which Apple (and Steve Jobs) have demonstrated allergies. That’s why Foxconn, not Apple, builds the iPhone. As a result, Apple has a $200 billion business, while carrying only $3 billion in inventory to support it. And that's not even speaking to the other up-front fixed costs invested by Foxconn to support Apple, which are passed on to Apple in the form of variable costs plus a (very small) margin.
Apple could buy / build as many chip foundries as they need and still have billions in cash stockpiled, if that is what they ultimately want or need to do.
You’re absolutely right.
Who is going to run these? Who is going to continue investing multiple billions on top of that, year after year, to develop the next process? And then spend another $10 billion three years later to build another one? If we’re talking about Apple buying or building fabs, the answer to all of the questions is “Apple”. And I don’t disagree that Apple has the money to do that.
But why would Apple do that? Especially when the main purpose (in this context) would be to build logic for Macs, a business that, as you’ve pointed out, is now dwarfed by Apple’s other businesses which don’t suffer the problems complained of about Intel because… Intel isn’t making those chips?
Just because Apple could do something doesn’t mean that it should (or will). Especially when doing that would require spending a huge amount of money, even for Apple, to do something that isn’t likely to provide a materially better solution or material growth to existing business over the status quo and isn’t likely to generate incremental revenue of its own (Apple investing in logic fabrication and then offering that technology to foundry customers? Yeah, no.).
At most, Apple might consider putting some lower end Mac or MacBook products on ARM logic of its own design, fabbed by TSMC or some other foundry. Maybe for a product along the lines of Microsoft’s Surface. Maybe. I mean, that’s not totally insane, but I don’t see the business or technology case for it.
But Apple getting into owning and operating fabs? Sell your shares. And then short sell more.
