“People who are really serious about software should make their own hardware.” — Alan Kay
Didn't help Sun Micro much :-). One could argue they weren't "serious" about the software I suppose.
This article is basically saying that Apple becomes nearly untouchable/catchable when they are building their own chips and devices and software. And yet that particular pattern has failed again and again. Digital Computer Corp was my first experience with it, Sun was my second, there are companies that decided to make Java "engines" which probably qualify as the third.
From what I have observed, companies and the software they produce can be viewed as exploring a graph where a local maxima is a successful product and a minima a failed product. If you can accept that model, then you can see that different hardware options can enable you to reach different maxima on the same graph, however once you've started building your own hardware you limit the spaces you can reach to only those where your hardware can reach them.
As a result, people who are trying to sell hardware (chip companies), being unconstrained by the local maxima you are pursuing, wander out into the solution space and uncover different and sometimes wildly better maxima that can solve the same problem. The vertical manufacturer then finds their bespoke hardware a drag on their ability to move from their current maxima to the one growing in the market adjacent to them.
If the new comers execute well they either put the vertically integrated old timers out of business or constrain their growth to a limited part of the graph.
I thought Apple had learned that lesson with PowerPC but perhaps they just felt they were doing it wrong and this time it will work. I'm guessing it won't work long term.
So far, Apple seems to be doing a good job of only doing their own hardware where they can turn that into a real advantage. They rightly gave up on the niche PowerPC architecture and switched to Intel CPUs. There are still tons of off-the-shelf components in Macs and even iOS devices, with lots of third-party IP in Apple's SoCs.
There aren't many examples of what looks like Apple doing their own hardware purely for the sake of being different and incompatible. Nothing stands out as a custom component that Apple would be better off replacing with a generic off-the-shelf part. And Apple isn't locking themselves into their own CPU architecture or anything like that; if they do end up falling behind with their mobile SoCs or something like that, it isn't hard for them to buy a different SoC with the same CPU instruction set.
There are really two different kinds of vertical integration.
One is where you actually integrate your products, making them less generic so they fit better together. So for example, if you design a CPU specifically to run Java bytecode then you make it more attractive to use that CPU (runs your Java code faster) and more attractive to use Java (runs faster on your Java CPU).
The problem there is that it ties you down. You end up with an ecosystem designed around the integration -- compilers for making efficient Java bytecode for that CPU, operating systems that put Java in low level places because that CPU enables it, etc.
Soon you have a house of cards and nothing can be changed or it all falls apart. When the market decides they want Rust and not Java then you can't sell your Java tools or your CPUs.
The other type of "integration" is where you just buy the good suppliers so your competitors can't have them, or develop quality generic components in house. Then you have the advantage of a commodity CPU which is a little bit faster than average, but if someone else comes up with an even better one you can still switch to it (at the rather expensive cost of writing off your previous investment).
The problem with that is your competitors can do the same thing and on average you're probably not going to be any better at it, so all you end up with is a series of bidding wars where you each buy up or recreate various pieces of the supply chain, and then internalize all the risk of an individual component under-performing the rest of the market.
The potential upside is that you may end up owning something that turns out to be really important and can be used to conquer the market, but that upside is dampened by the fact that if it actually works you'll soon have antitrust lawyers from six different countries at your door. And it also tends to have a long-term destabilizing effect on the business, because temporarily having high margins and no competition simultaneously allows the company to become complacent and encourages third party investment into disrupting the market. It's a strategy designed to produce a trillion dollars in revenue for a year rather than a hundred billion dollars every year for a century, at a nontrivial risk of taking down the whole company without first conquering the market.
A lot of this is really down to the principal-agent problem and human biases. Each person thinks they can do better than the whole rest of the world.
So when you have a large company producing more profits than it can productively reinvest in its own line of business, rather than returning the money to the investors so they can choose what other business to invest in, the temptation is to expand into another market. Because if you're successful at one thing you should be successful at every other thing, right?
So you get a huge conglomerate that looks good on paper, but unless the managers have better investment acumen than the stock market as a whole (or are very lucky) what they're really doing is burning the investors' money on below-market investments and hiding the below-market returns in the above-market returns from their primary line of business.
It's also becoming increasingly difficult to dampen this because nobody actually owns these companies in the sense of being able to affect management. Even if I had a million dollars of AAPL shares, if I don't like what the management is doing, what can I do? I can vote, but I have ~0.000% of the shares. I can sell my shares, but that doesn't change their behavior. They're still outperforming the market (until they don't), even if the alternative would be even better performance, so no majority is inclined to oust them until after the damage is already done and proved. So what's the check on wasteful empire building?
Or does the size of these companies prove that there isn't one.
Interesting so far no one has put out any numbers yet.
There is one important factor that is missing in your thesis, unit economics. None of the companies you names, Power PC Apple, Digital Computer and Sun are are anywhere near the unit economics Apple is currently holding.
Out of the 1.3B Smartphone sold yearly, Apple has 220M, ~17% of Total Market. With ~40M iPad, Apple TV, HomePod, Apple is approaching 300M SoC that are produced for the iPhone ( I am ignoring Apple Watch S1 for this reason which has a similar design for its own ). This number is bigger than the 260M microprocessor unit Intel ship every year. ( While some people likes Horace Dediu from Asymco like to mention this, I would also like to add the number from Intel does not include Chipset, and SSD Controller shipment, both are included in Apple's calculation. So not an entirely fair comparison, but a relevant one )
When you command a substantial percentage of the market, and more importantly, a near monopoly in the premium market, it makes sense because the unit economics not only saves you money, but also allow you to innovate on your own pace. The Apple W1 / W2 for example are one the best low power bluetooth / WiFi chip, instead of paying EXTRA $5 per chip for another company to sustain their OPex and R&D, they are doing the work themselves which will later filter upwards to those 300M iOS unit, or Broadcomm will try to please Apple with a price that Apple thought is not worth doing it on its won.
And it is precisely the same reason why I don't think Apple will switch to ARM on the Mac. The Mac CPU list consist of TDP from 6W to 220W, 4-5 different TDP variation in between those 25M unit. That is anywhere between 20 - 40x difference in unit economics.
Although Apple being Apple, I believe unit economy is only one driving factor. The first reason they make such move would be they are not getting what they wanted, and if it is not good enough they will decide to make it themselves. Saving money was only ever the consequence, not the reason.
All your examples are "people who are really serious about hardware" - not software.
A better example is perhaps the British company Acorn, which made computers for a couple of decades but ultimately morphed into the very successful company ARM. (Which originally stood for Acorn RISC Machine.)
Their software was eventually, after the failed Oracle NC project and a number of legal shenanigans, nearly-open-sourced as RISC OS Open, and has a small number of mainly British dedicated followers, but otherwise did not make a huge dent in the universe. https://www.riscosopen.org/
I think you are drawing the wrong conclusions from history because you are conflating platforms and companies.
People think about the PC as successful because the platform is so big. Yet for the company that created the PC it was a failure. IBM lost the PC battle.
Sun would have lost whether they had made the software or the hardware, because the platform they made hardware and software for was outcompeted.
And anyway I think there is a significant difference. Sun and SGI were essentially hardware companies which happened to make software. I feel Apple is the reverse. It is a software company making hardware.
Apple harware was never impressive, sun harware was. Apple was all about the experience the software gave you. The hardware was just built around the software to facilitate that.
They have to keep as far away from being another HP or Dell as possible as more and more services head to the cloud.
I look at Apple as being like EMC when they were printing cash. EMC was selling high margin software with hardware. You paid maintenance on both and re-purchased the software every refresh.
The question with Apple is how sticky the services are. Those blue iMessage bubbles are powerful peer pressure. I don’t think that Sun, DEC, or EMC had anything like it.
The other thing I like about Apple is they are less exposed to political winds of change. Government (PRC) have already broken Google in their locales. Apple is not as vulnerable imo.
Something like 90% of EMC kit is off the shelf. Their CEO said so at some point. That other 10% is the key. That is what Apple is doing so well. Marry the HW to the software where and when needed to enable a feature/function better then the rest.
Not only the hardware components - the lengths Apple goes to differentiate are nothing short of astonishing - from heat removal, to structural strength, to pioneering keyboard switch designs, while it's fair to say ThinkPads offer a better value for the money, you can't say Apple machines don't seem to come from the future.
What killed SPARC, Alpha, and MIPS wasn't that they became worse than others, but that cheaper x86 boxes running Linux became good enough. There is a point where running your workloads on higher end hardware ceases to provide benefit and only cause lower hardware utilization.
I'd love to be able to justify an IBM z14 or one of the higher end Sun (I refuse to call them Oracle) boxes, or even an 8-way Xeon Platinum, but more often than not, a couple VMs on your favorite cloud provider will solve the problem neatly with simpler accounting maneuvers.
Apple is already using its own IP on iPhones and iPads (and, increasingly, in components for their x86 computers). They also have a good idea of what their hardware runs and that places them in a great position to optimize the hardware to their software - to the point they could implement OS-level functions into microcode or do some other types of highly specialized tuning other companies who make chips for the general market can't (or won't - see the PPC heat issue that made Apple go Intel).
That said, Apple didn't abandon PPC because PPC couldn't be better than Intel. They abandoned PPC because their laptop volume didn't justify the investment needed to make PPC G5 have similar performance per watt to what Intel was achieving with their Core series (itself a direction change from the Pentium IV family's dead end). IBM offered Cell to them and Jobs didn't think it'd work (the Cell, with its crappy PPUs and memory-disconnected SPUs is probably the hardest thing to program I've ever seen) well. Intel, on the other hand, was looking for a huge PR win and Apple was it.
Sun, DEC and SGI made the same mistakes imho; they ignored PCs for far too long. Apple too by the way. They thought people wanted custom hardware and software to render movie, run 3d software, do scientific computing etc and for a while that was very true (as someone who has many Sun, DEC and SGI machines still at home and they all still work perfectly, which you cannot say for PCs from that age, for me it is still true but alas), but PC hardware, software and economies of scale passed them by and they responded far too late. It went the same way with mainframes to minis, minis to workstations, home computers to pcs, so it was not like they did not have examples of this in their past. It was printing money in the 90s for them though, things were just too good.
I certainly agree with this, but both Sun and DEC started out very 'open' and then got more and more closed and vertical and then they died. I have a couple of custom chips in my PDP-8, perhaps a dozen in my PDP-11, and a whole bunch in the VAX. Similar pattern with Sun-2, Sun-3, Sun4, and then Sun 10/20/Ultra... It was the very success that they were experiencing that lead them to believe that they could capture more and more of the 'value chain' until they couldn't any more. I expect this will end badly for Apple but I am open to other ideas about why it won't.
I think one difference is that Sun and DEC were selling to enterprise/business markets whereas Apple sells to the consumer market. I think the former tend to make decisions more based on cost/benefit while the latter tend to be more emotionally driven. If Apple can continue to succeed with that I think they will continue to survive.
What killed Sun and DEC and other RISC manufacturers (also PowerPC) was, that once the PC market grew to a certain size, they couldn't keep up with Intel as Intel could afford a magnitude more R&D costs with their processors and equally had lesser production costs. This shows, that going vertical won't work, if your production numbers drop below a certain threshhold, the economy of scale would work against you.
For Apple, the combined range of iOS devices currently gives them enough scale.
I think Sun could have had a much better chance, if they had pushed SPARC processors into the private market too, like offering a decent SPARC based ATX motherboard for all the PC enthusiasts setting up their Linux machines.
I don’t think either Sun or DEC started out as very open, at least in hardware, and probably in software either. Much of their openness comes at the end of their reigns. At any rate, I don’t think this had very much to do with their downfalls.
In the 1960s and 1970s, every DEC computer came with a full schematic. You could troubleshoot any signal on any board. Most of it was COTS components, and replacements were easily obtained. Also, by 1980 there was a thriving industry of 3rd party add-on boards.
The operating systems were less open, but that had a lot to do with the cost of distributing megabytes of source code. It was possible and recommended to recompile RT-11, at least, from source to optimize for your hardware configuration.
The difference between PowerPC and Apple's A series is compatibility. If somebody comes out with ARM-based SOC that is leaps and bounds better than what Apple's got, they can easily become a customer and build an iPhone around that rather than their own work. There would be no need for a long and difficult transition like the one from PowerPC to Intel. Another option would be to buy or license specific components and incorporate them in to the next A-series SOC. The haven't boxed themselves in at all.
As for compatibility, Apple managed exceedingly well the moves from 68K to PPC and then to Intel, even keeping some binary compatibility. Few companies ever did that - DEC comes to mind, making VMS and Ultrix move from VAX to Alpha, and IBM moving the iSeries from AS/400 to POWER. iOS software will continue to be written for ARM for as long as we can see. macOS can, eventually, move to ARM too - if Microsoft can do it (emulate x86 at Celeron speeds on Snapdragon), Apple certainly can.
While the compatibility certainly eased the transition path for Apple, I think the main differentiator between PowerPC Macs and the Apple-ARM iPhones is the number of units. The PowerPC machines never reached the unit count allowing for a sustainable development budget to keep up with Intels x86 investments.
With the iPhone, Apple has the production numbers (and especially the revenue) to match or exceed the development budget of any competitor in the mobile phone space. It also helps a lot, that currently the best processes are available from independant fabs, which was not the case in the past.
Nah, it’s simpler than that - once you bring a dependency in-house, it’s almost impossible to shop for the best option on the market because your internal product is always “the best” and if you don’t agree, then you are a “jerk”.
It's not just the politics, it legitimately changes your whole cost structure. If you spend ten billion dollars in R&D and then someone else comes out with something that is just better, you have a big problem.
You still need a hundred million units. The external supplier charges $200/unit. You paid ten billion dollars so you could manufacture your own at an incremental cost of $100/unit. Are you going to pay another twenty billion dollars (total of thirty) to buy all your units from them, even though your competitors are paying twenty in total? Or are you going to carry on manufacturing your own to contain costs, but then your competitors have a superior product? Either way it's a major competitive disadvantage.
Apple doesn't make their own hardware. What they do is design their own hardware and chips. I agree with you that if they actually made their chips this would be bad.
Case in point, you can see Intel really struggling because not only do they design their own hardware but they also make it quite literally with their own fabs. This has resulted in a lot of delays moving to 10nm and beyond.
AMD on the other hand spun out their fabs and now use fabs around the world to handle the actual manufacturing. This frees them up to actually design hardware and innovate at a much greater clip. Heh, they are even going to 7nm.
The same applies with Apple. It is quite cheap to be vertically integrated because you simply don't really have to make anything. The capital investment is quite low. It is something that Sun didn't have the luxury of.
The problem with quoting process nodes from different fabs is that there isn't any standard for defining the minimum feature size. IIRC, the gap between the process node's nominal and the actual minimum feature size began diverging after the 45nm era and has devolved into being absolutely meaningless at this point. Personally, I believe the vertical integration afforded by owning your own fabs actually pans out even though Intel appears to be struggling at the moment. AMD divesting their fabs was one of the largest reasons why they were so uncompetitive up until recently and a large portion of their current renaissance lies in the fact that we have reached the end of the line in terms of consistent process shrinks. There's only a few nm left until we arrive at the physical barrier of individual atoms and the independent fabs have now had enough time to catcb up to the very frontier of process engineering.
At JuliaCon 2018 in London, Sophie Wilson (Broadcom/ARM/Cambridge) gave an insightful overview about the "Future of Microprocessors", and she touched on this topic:
Having your own fabs is actually a huge advantage. It gave them so much advantage that they got lazy in the last decade (in the absence of serious competition).
It seems like what you're saying is less that vertical integration is good (which you're actually arguing against) and more that Apple isn't actually that vertically integrated.
What I am stating is that the definition of being vertically integrated has changed. The OP is arguing it is bad because of the history of capital costs associated with manufacturing.
It is not longer cost prohibitive to be vertically integrated because you don't have the upfront capital costs of manufacturing. With that, Apple is excelling at what they are doing. AMD is excelling at it as well. Historically, this type of behavior would be risky if not foolish.
However, companies need not be constrained to the local maxima you describe.
They can switch hardware platforms. In fact, Apple is the obvious successful example of this, where the Mac line has moved from Motorola to PowerPC to x86/x84. Not to mention, the introduction of an historically successful new product line on ARM.
The general problem with hardware platforms is that they have very high fixed start-up costs. This is exacerbated by the fact that they are moving quickly so that these start-up costs are incurred again and again for products on the forefront. The way to deal with this economically is have large scale. Usually that scale is achieved though a market in which multiple companies are the buyers. But in Apple's case they are apparently large enough to achieve that scale by themselves for the iOS lines so are able to bring that in-house and then get the additional advantage of control over the hardware.
Now, of cours, Apple's A-series SOCs may well reach a maxima at some point in the future, and then get bypassed by some other hardware technology (though it's not like the A-series is sitting still). That doesn't mean Apple can't or won't adopt that new platform. And if they don't, that has nothing to do with whether their old platform was in-house or out-of-house.)
In general, this is about gaining control over key technologies they rely on due to their scale and strategic planning, which is generally a good thing. Of course, they also need to execute, but that's true entirely aside from "going vertical".
A key difference is economy of scale. Apple produces a much larger number of chips than Sun’s or DEC’s markets could support back then.
Some other specialized vendors may still produce more chips but their scale advantage over Apple’s is relatively minimal and Apple’s ability to customize chips for its own platforms may outweigh that.
Dont think I can agree. Pretty much NVIDIA took over the batoon fron them during race unless they do something more that graphic design.
Problem with SGI in large part they kept price of their hardware very high and focused only on large enterprises. Not long after otheir graphic card designers came around and even if power was smaller you hust took few at the same time and it made up. Costwise SGI was tenfold more expensive.
NVIDIA is verry narrow with their product and I dont see them going anywhere or sharing SGI’s fate.
Not really sure what this article's final point is? Facebook and Google have hardware teams that allow them to build impressive data centers. Cisco has lost a lot of potential revenue from all the components FB and Google have managed to build in-house.
Even if you say that's not the same, Samsung is probably the closest to Apple in the consumer space. They own the entire pipeline in various degrees.
Yea Apple is winning in profits, but it's not the vertical integration that's doing it.
Less and less every year. Five years ago the entire DC used third-party networking gear from top to bottom. The top of rack switch was the first thing to go, replaced with an in-house device that could do the job well and that provided flexibility for future plans. Load balancers (there was almost a contest to see who could get to be the ones to take a sledgehammer to those F5 pieces of shit) and cluster switches were next; at that point you had an entire fabric of custom-made kit that would do rack-level BGP. Intra-cluster and gateway switches are the remaining big iron, or at least were when I left, and I am not sure it will ever be worthwhile for Facebook to invest in the hardware necessary for those tasks. Expensive kit for sure, but the spend with third-party networking gear companies dropped by at least an order of magnitude over less than a decade.
Their description of PA Semi is a little odd. Yes, I'm sure there was some ARM work but at the time they were far better known for PWRficient and the PA6T, and they still had military contracts for it at the time they were acquired. There were rumours that was going to be the next PowerBook chip until Apple went Intel.
Thanks, I though I was the only one that remembered it still.
I was salivating for an Apple designed PPC back then and was devastated when they switched to Intel.
Even though it was probably the right decision, the world got a little more boring with effectively a single ISA for mobile and desktop.
Same with OSs, something died with the original classic OS that never recovered, even though pragmatically, I love being able to run Photoshop on laptop Unix machine.
Yes, I remember the feeling well when the announcement came out. And I agree that Apple's human interface zenith was sometime in the classic Mac OS timeframe.
Oh, well, I've got a Talos now. I don't suffer any illusions that it'll bring Power ISA back to the desktop in any great numbers, but it's at least competitive.
From the article: Steve Jobs quotes Alan Kay at iPhone 7 launch.
Announce date of iPhone 7: iPhone 7 and iPhone 7 Plus are smartphones designed, developed, and marketed by Apple Inc. They were announced on September 7, 2016 (https://en.wikipedia.org/wiki/IPhone_7)
But the question is: should developers outside of Apple applaud this trend? How about consumers? Or small startups? Will complete vertical integration eventually spoil the market?
I don’t think there is too much of a market to spoil. Android manufacturers and PC makers who are just using commodity off the shelf undifferentiated hardware and software operate on razor thin margins in a race to the bottom.
Eh, do not really care. All of my apple kit just works. It gets out of the way and let’s me do want I need. As long as that is the case then it works for me, and most others.
Apple could buy a fab but doesn't, so vertical integration isn't a religion for them. Nobody said it was, but I like pointing out their restraint. Maybe when the semiconductor that replaces silicon, say diamond, is close to viable, they'll splurge, not wait for it. They were worried they wouldn't be able to differentiate the iPhone, that's why they bothered with the A series chips. Forward thinking led them to see other phones improving beyond "good enough" or whatever. Maybe VR was on their horizon, too, or something else unknown.
I like to imagine the satisfaction felt by whoever got the board of directors to approve going the direction of designing chips, not that they couldn't afford an R&D folly. Everybody likes to tell Apple how to spend its money, but almost everyone is wrong. Whoever championed ARM acquisitions must feel extra smart, since they were proved right. But will Apple ever make a super inexpensive notebook or PC powered by an A series chip? Probably not, because of stupid economics. They could, but they won't.
Ok you said they didn’t buy a fab and you weren’t correct
Building and operating a fab for 7nm is a massive undertaking.
I find your comment incredibly naive. It cost 10s of billions of dollars to develop just the 7nm process alone. Global Foundries just cancelled all development of 7nm and beyond to focus on specialty processes and their 14nm process. Building an advanced node Fab would be an enormous undertaking and take many years before the first production wafer came out. Not to mention that there’s no guarantee that fab would be successful. On top of that fabs are all about volume, they are hugely expensive to operate. So it’s unlikely an Apple only fab would be economically competitive.
It was started by ex-Digital employees that previously worked on the StrongARM, thought that is still an odd mention since IIRC they worked on MIPS cores in the time between.
“People who are really serious about software should make their own hardware.” — Alan Kay
Didn't help Sun Micro much :-). One could argue they weren't "serious" about the software I suppose.
This article is basically saying that Apple becomes nearly untouchable/catchable when they are building their own chips and devices and software. And yet that particular pattern has failed again and again. Digital Computer Corp was my first experience with it, Sun was my second, there are companies that decided to make Java "engines" which probably qualify as the third.
From what I have observed, companies and the software they produce can be viewed as exploring a graph where a local maxima is a successful product and a minima a failed product. If you can accept that model, then you can see that different hardware options can enable you to reach different maxima on the same graph, however once you've started building your own hardware you limit the spaces you can reach to only those where your hardware can reach them.
As a result, people who are trying to sell hardware (chip companies), being unconstrained by the local maxima you are pursuing, wander out into the solution space and uncover different and sometimes wildly better maxima that can solve the same problem. The vertical manufacturer then finds their bespoke hardware a drag on their ability to move from their current maxima to the one growing in the market adjacent to them.
If the new comers execute well they either put the vertically integrated old timers out of business or constrain their growth to a limited part of the graph.
I thought Apple had learned that lesson with PowerPC but perhaps they just felt they were doing it wrong and this time it will work. I'm guessing it won't work long term.