How many iOS devices will be produced in the next 12 months?

Apple’s capital expenditures for product tooling, manufacturing process equipment, corporate facilities and infrastructure has followed very closely their production of iOS devices. The pattern is shown in the graph below.

Screen Shot 2013-10-30 at 10-30-2.12.28 PM

Note that although reported expenditures did not match forecasts for 2012 and 2013, the differences nearly cancel each other.[1] The company’s forecast for fiscal 2014 is shown as well.

The orange line shows iOS unit production[2] with the scale on the right-side axis. Note the correlation with forecasts on CapEx. The relationship can be seen more clearly in the following scatter plot.

Screen Shot 2013-10-30 at 10-30-2.33.26 PM

I added the 2014 Forecast ($10.5 billion from the latest 10-K filing). If the relationship holds into next year then the iOS unit shipments should be between 250 million and 285 million.[3].

  1. I suspect that the difference might be caused by a payment being brought forward in 2012. []
  2. Technically, shipments []
  3. The equation for the trend line shown suggests 285,908,000 units []
  • Walt French

    This connection looks like as good a way of projecting Apple’s expected shipments as any.

    Yes, there are all sorts of timing differences — whatever happened with the supposed near-term hookup with TSMC? — and possible bottlenecks such as many people read into the comments on availability of the Retina iPad Mini. But it seems like the fit here is pretty good.

    There’s a bit of a “model specification” concern, though. If CapEx was fully exhausted by the next year’s production—inside of 12 months—it wouldn’t really be a capital expenditure. Those machines that milled the 4S cases should have been repurposeable for 5 and 5s cases; ditto factories full of assembly lines to stuff boards into the tiny cases, etc. Maybe a machine is so product-specific that it’s written down to zero after a year, but more likely, a dollar of CapEx is put to work for 3–5 years.

    That means we shouldn’t be looking at one year’s CapEx versus just the next year’s sales. I.e., if next year’s iPhone6 and iPad Pro use similar formfactors and technology, then CapEx should only be necessary for incremental sales on top of this year’s production. Unless Apple is bringing production inhouse in place of buying subassemblies that it used to get from eg CPU suppliers or case manufacturers, the CapEx should be driving increases, not units per year.

    The nice fit may be an artifact of the strong trend increases underneath units sold. (“De-trending” is a common modeling technique to find the timing relationships where lifetimes of inputs are unclear.) If iOS sales are growing more slowly, subsequent CapEx should be for new processes, more inhouse work, etc.

    The new Austin facility doesn’t seem to have enough product demand to require huge amounts of CapEx. If there ARE large amounts of CapEx there, we should expect more than a few hundred thousand or low millions of MacPros to roll out from it, and the CapEx costs will have to be justified against the costs & control of the very reliable and well-known contract manufacturers in the US and elsewhere.

    Alternatively, if Apple doesn’t want to own its own foundry business, and doesn’t have major new product categories underway, we should see CapEx drop. Until we know more about the nature of the CapEx, we’ll be left with iOS forecasts that are less certain than the historical fit suggests.

    • johnnygo

      I agree that one has to take any projection based on capex with caution as there are many different issues involved at the same time.

      For example, investment for the new iPad Air line should be significant as it is a totally different product from the iPad 4 (casing, GF2 glass, etc) but at the same time, it shares similar manufacturing with the iPad mini and the iP5/5S (diamond shafers, similar case to iPad mini, etc).

      Other products, like the iP5S would require minimal capex as it seems to be produced entirely on the same process as the iP5 with any capex going only for additional unit sales.

      I suspect the new campus has a significant impact both in the 2013 figure as well as in the 2014 expenditure. Dito for the data centers Apple has been building over the last 2 years. Those are clearly not production related capex.

      Another issue is Apple move away from Samsung. Be it on displays or in chips, I suspect this strategic decision is increasing capex by $1-2 billion per year without any corresponding unit sale increase (but I agree that there is significant strategic value to pursue this strategy).

      • There is no impact in capex from the new campus. Land was purchased in 2010 and there has been no construction yet.

      • John

        But might there be in the 2014 $10.5 billion, and consequently the 285 million units?

      • Of course. The only question is how much will the project cost and how quickly are those costs incurred. There are also subtleties in which costs can be applied.
        The project’s time frame and acceleration of effort could vary and the impact would therefore easily fit within the margin of error in the capex estimates that we’ve already seen.

    • nuttmedia

      This is an excellent point. Many unknown variables. A significant public one is the cost of the new Apple HQ. With estimates ranging $3-$5b and a 2016 move-in date, you have to assume a decent chunk in the 2014P figure.

      That said, there is FP&A convention that gears capex to contracted production and such. Lots of noise, to be sure, but as you said, it’s as good a tool as any other.

      • Construction costs are not entirely capex. But even assuming they are, the cost of construction will be around $1 billion/yr though obviously not linearly. Land purchase cost was allocated in 2010.

    • Javbw

      As I understand it, the bulk of the machine tooling orders is for specialized machines for making specific parts for a specific model. I read somewhere (from Horace?) that they are depreciated over 2 years.

      But the machines themselves are a cog in the machine they make to build phones. For example, the milling machines (and other specialty metal shaping tools) used to create iPhone 4 and 4S’s are either depreciated and broken, or still in the remaining line producing the 4s and 4S models still being sold today. Working 24/7 for a few years probably wears out these milling machines pretty fast.

      Same with the 5 and 5S – they are working away.

      Next there was 5C machines (you don’t need as many) and the new mini case doesn’t look stamped, it looks milled, so more milling machines.

      With the machines reaching end of production life quicker than we might expect – this is probably the reason others don’t use them as the ROI on them is lower than simpler machines doing stamping and plastic molding – Apple is footing the bill because they are expensive, but a signature way to separate them from the pack – and they have the margins and scale to hide the cost in.

      • Tatil_S

        In FY 2012 and 2013, Apple spent 17 billion dollars on CapEx (I am trying to add back the expenses related to the stores) and took out 10 billion dollars of depreciation. Net increase in PP&E during that period is 8.8 billion. Hmm, it does not quite add up.

        In FY2013 only, 7 billion of spending on CapEx and 6.8 billion of depreciation results in 1.1 billion net increase in PP&E. Again, my simple calculation is off, but overall it seems to support your thesis that most of the equipment gets depreciated very fast, making them almost a variable cost rather than a fixed one.

      • Walt French

        Just as a thought exercise, if it *were* true that CapEx was economically run down to zero after two years—and if also, Apple were not taking new processes under its wing, whether at FoxConn or its own factories—then you’d want to model sales over the next two years as a function of a prior year’s CapEx.

        There are some subtle modeling issues in overlapping time periods, and what with limited number of years (and seasonality making quarterly data less useful) I don’t think you’ll get much better insight for the amount of work.

        But the point still stands. Just as every satisfied customer is a likely repeat sale (or a “halo effect” sale) in a couple of years, CapEx is at least somewhat cumulative in its effect on sales.

        PS: this effect implicitly incorporates Apple’s predicting its sales precisely, an impossibility to which Apple comes very close while not very often being supply-constrained for long.

      • tmay

        Lets say that these machines are running around the clock; 8750 some hours per year. Depreciation in two years under those conditions might be reasonable, but there is plenty of value left in the machines at that point.

        Wear parts are bearings, linear guide, ball lead screws, spindle assembly, relays. All but the linear guides for the x/y axes tables and z axes travel are relatively easy to replace, and life of a linear guide is a factor of lifetime travel, acceleration and table, fixture, workpiece mass and the mass of the z axis structure. I’ve seen data indicating 1200 km of travel as being a nominal service life.

        Apple has probably looked at the x/y/z travels for typical families of parts, and derives an expected lifespan for the machine tool, but also has an option that might take the machines out of service at a certain hour limit, renovates them with replacement wear parts and repurposes them for current production requirements. Not sure how the financials work out, but it would seem logical for an ever expanding workload.

      • steve

        excellent point. In many manufacturing industries it is common to buy a service contract for a piece of machinery and extend its useful life. Very common in the automobile industry. These contracts are usually about 10 – 15% the purchase price of the tool per year.

  • Chris

    I might be missing something, but are the iOS units shipped your own calculations? Units sales by product show around 220,000k units sold (iPhone and iPad), whereas the chart shows above 225,000k. Is this due to the Apple TV?

    • Chris

      iPod touch and AppleTV? Are these both your estimates, excluding some years when management have disclosed figures?

      • iPod touch estimates are based on the “more than half” of iPod as reported by the company. My estimates are about 55% of iPod on average.

    • iOS includes iPod touch. I do not include Apple TV.

  • markrogo

    This is fascinating as usual. 285 million sounds like >almost< good news for Apple, though not quite. If we presume significant iPad growth to 90 million and 5 million for "other" that leaves 190 for iPhone. That's a <30% bump over the prior year. It's not a disaster, but it will leave Apple in the low double digits of the smartphone market and raise the calls for adjustments to iPhone pricing yet again.

    • iObserver

      I caution taking market share data too literally. It is based off actual numbers given by Apple and a schmorgas board of guesstimated unit shipments from other manufacturers combined with not-100%-reliable internet usage data (the latter, by the way, would suggest Apple has a much higher smartphone share).

      • neutrino23

        Market share is an interesting thing to consider, but full of problems. In addition to the shipments issue, you have to consider the very fuzzy concept of what constitutes a smart phone or a tablet.

    • charly

      Significant iPad growth?

      Apple is lucky that ipad has been stable this year. There is no growth in ipad numbers. Not surprising with their much higher cost and an operating system that isn’t suited to be used as office tool

      • twilightmoon

        Not suited to be an office tool? Am I the only one who thinks this tired argument has the stench of pure trolling? I’m going to assume that you are not a paid astroturfer so please justify your statement.

      • Gary

        It’s not meant to be office tool

      • jpintobks

        I disagree charley. I manage to run a small publishing company (more that 130 print and eBook titles) with an iPad. Invoicing, production, copyrights payments, communication, design, distribution and editing. It is an incredible business tool

      • Walt French

        I think you missed a little point: in the past 5 years, business purchases of all PCs, as well as consumer purchases of Windows PCs, have stagnated and started falling significantly.

        That market does not need new machines every year; my old HP desktop keeps working fine but neither it nor an Ultrabook works well to take the dozens of PDFs to the meetings I attend, nor to read my twitter feed with the AM news, on the train.

        New uses; most don’t require my monstrous spreadsheets or the 60-slide decks my marketing dept builds. So companies are increasingly looking to more appropriate devices. Individuals are looking for something quick & easy for home/travel. Students something that has dedicated apps (and no they do NOT need a chiclet keyboard to be highly productive). Kids something that’s fun to play with and to take in the car.

        It goes on. The world has a much bigger definition of “useful” than you paint.

      • Space Gorilla

        Well, iPad unit sales by fiscal year are:

        2010: 7.5 million
        2011: 32 million
        2012: 58 million
        2013: 71 million

        iPad numbers are in fact going up.

      • Space Gorilla

        And on the subject of iPad numbers, charly was never heard from again 🙂

      • charly

        12Q3 to 13Q3. What were those numbers?

      • Space Gorilla

        Sorry, cherry picking data to fit your narrative won’t fly. Come on, you’re not really this dumb, don’t be that guy. iPad unit sales continue to increase year over year, that’s the simple truth. Quarterly snapshots of a cyclical product aren’t all that useful (it doesn’t prove what you wish it proved). What I find interesting is that the iPad broke 70 million in annual sales in four years while it took the iPhone five years to break 70 million (roughly speaking). That might tell us something about the future of the iPad, given where the iPhone is now.

      • charly

        It is not cherry picking. Ipad exploded but the last quarter was stable y-o-y, that is significant. Especially when back to school is IIRC in Q3.

    • I’m sure that 285 million iOS units sold will be seen as a disaster by some. It is, after all, only the size of the PC market.

    • Space Gorilla

      To get to 90 million annual iPad sales Apple needs roughly a 26 percent increase in sales from 2013 fiscal year. Not sure that qualifies as significant, since you go on to say that a similar increase in iPhone sales isn’t very good.

      • JohnDoey

        The iPad Air and iPad mini with Retina lineup could be considered a reboot of the line. They are both equally high-end and could both be said to be next-generation tablets, with ARM64 and Retina and low-power Bluetooth opening up new applications. I think we will see really great iPad sales in 2014. There is a reason to upgrade even if you have an iPad 4 or original iPad mini. And at the same time, an almost total lack of tablet PC app development for any tablet other than iPad.

  • Looks like a log fit to me.

    • Micromeme

      Me too. Or rather whats the projection if you assume thats a logistic curve we are looking at

      • Ok the first chart could be logistic in time but I don’t think it would affect the correlation. The time-series could be anything and the regression/scatter plot still be linear or some other special function.

        What I meant was, it looks like a logarithmic fit on the scatter plot: take the logarithms of the x variable (Forecast CapEx) and use that for the linear regression/scatter plot. Or if on Excel, just choose that type of trendline.

    • I tried all the basic fits. Polynomial is much closer than log but not as close as linear. Of course you can throw away a few points and get different results. There are not enough point to make this worth while. The point of this analysis is however that capex *should* be a direct function of output since it’s mostly tooling. To assume otherwise would require some complicated arguments: that somehow Apple needs to spend more for less or equal output as the product matures.

      • “To assume otherwise would require some complicated arguments: that somehow Apple needs to spend more for less or equal output as the product matures.”

        Diminishing marginal utility/returns might be a “complicated argument”, but it’s a widespread economic concept.

      • Walt French

        Indeed, a major textbook reason why companies can be limited in size. But the explosion of global trade, especially the phenomenon of ultra-huge factories in China, plus the “positive externalities / network effects” of concentrating production, suggest that increasing returns to scale are likely.

        (Such are the recent thoughts of an economist who’s won a Nobel prize, about trade in the 21st century.)

        Think about a factory in China with 3 X machines, considering a 4th. Very likely, it’s more efficient to add in that 4th machine there, rather than in Sandusky, Ohio. More efficient for the COO to be sure that the parts inventories are in their correct location, etc. I think those types of examples betterdescribe what Apple faces, than the expenses of trying to jam too many things into a fixed space.

      • That may well be a sound theory, but how does it explain the empirical result shown in the scatter plot of clearly diminishing marginal returns (except perhaps for the first 3 or 4 points)?

      • Walt French

        I don’t think that Horace *is* claiming (nor should he) that Apple’s capex decisions ONLY reflect a constant ratio to labor, purchased subassemblies and machinery. We *DO* know that over the years Apple has made not just up-front purchase commitments, but also started purchasing equipment for use outside of Apple. We also know that in the last three years, Apple is making fewer changes to the basic design of its devices, which would favor shifting from manual labor to mechanized assembly. That’d reduce overall expenses at the same time output per $ of capex went down.

        More generally, there are too many unknown influences on capex and its utilization to take these few points as evidence of either increasing or decreasing returns to scale. Indeed, you’d see “decreasing returns” if Apple simply over-estimated the number of machines needed to meet demand.. I think that’s a poor explanation, but it indicates how little we know. As I wrote elsewhere, this is a useful first cut and I think we should let it go at that… we’ll never have the details to refine the model much.

      • Walt, sorry but I’m not following much of your first paragraph. Not understanding your arguments there. Constant ratio of capex to labor and machineries…? I didn’t say Horace was claiming any of that, I don’t think? Purchase commitments, device design, and mechanized v. manual labor? How does all that relate to the observed concavity of the curve made by the points on this particular scatter plot? I think it’s a good thing that much of classical economic theory supports a concave curve for this scatter plot. Looks a lot like this:

        The convexity in the very first few points of the scatter plot does reflect the very valid arguments for increasing returns to scale. That was 2007-2010. But 2009-2013 looks bent the other way. Nothing wrong or complicated about that, as it fits classical economic theory of production.

        I do agree in general with your second paragraph.

      • Seven data points. There is too little data to make conclusions about curves from such little data. Not only that but the data is not representing a perfect proxy. We have multiple components that make up the data such as the data centers being built and the differences in product tooling demands and partnerships, etc. It’s very dangerous to read too much false precision into this. I tried to do it on a quarterly basis and it was a disaster since there was vast shifting of spending due to random reasons.

      • Walt French

        Imagine some function — say, moving a given subassembly from one machine to another — can be done by either a robotic machine costing $1MM (for which a fair rental, considering a 4-year likely lifetime, is $200K) or by 8 workers at $25K/year each.

        To a first approximation, Apple is indifferent between pthe three choices of having Foxconn hire the workers, having Foxconn buy the machine (either of which increases expenses by $200K) and buying the machine for Foxconn as $1MM capex and getting the same payback in lower production expenses.

        Three different categories in accounting, however, even though the outcome is the same.

        It’s the latter category that Apple seems to be pursuing. Perhaps by having control of the design, they can better project the future utilization (and the $1MM machine will only cost them $150K/year, or because their cost of capital is less than what FoxConn has to pay when IT borrows, it’s seemingly been advantageous for Apple to buy the capital equpment.

        With relatively little impact on production, making it look like diminishing returns to capital because we’re not looking at the increased returns to every $1 of labor employed (same production, fewer bodies).

      • Perhaps I’m naiive but marginal utility seems a complicated argument to make for production tooling. Their value is not a matter of perception or relative pricing.

      • Why not? Who creates these tools and machines and factories and whatever other things that may be involved, and provides them to Apple? Is there no perceptive entity involved in pricing these, so that it is economically sound for them to make them at such quantities and prices? Isn’t Tim Cook weighing relative prices and quantities in a market of these things in order to procure Apple with enough at a financially viable cost?

        Aren’t such things made up of physical materials? Aren’t they designed and created by skilled, specialized labor? Aren’t these, at some point, potentially scarce resources, subject to laws of supply and demand?

        I imagine if Apple, likely the largest purchaser of these things already, were to require an order of magnitude increase in output, they’d have to pay more than a similar order of magnitude worth of tooling and equipment, as prices for these most likely would increase. No?

        Seems logical to me, but perhaps I’m naive as well. Maybe some of the other, more knowledgeable readers could chime-in since neither of us is an economist (tongue in cheek).

      • What you are suggesting then is that the price of the tooling goes up even though its utility doesn’t. I suppose that’s possible. I had not thought of it that way.

      • “spend more for less or equal output as the product matures”

        Not spend more for less or equal output, but for a smaller increase in output relative to the increase in spending.

      • “Polynomial is much closer than log but not as close as linear.”

        I went ahead and tried it myself (with my own estimates for iPod touch, but I doubt those make any difference), and found the polynomial a much better fit than linear (R^2 of 99.5% v. 96.8%).

  • How does construction of the new headquarters fit into these CapEx projections? Building was supposed to begin in 2013, but was delayed to 2014. Could the $3 billion shortfall in actual CapEx expenditures in 2013 (compared with forecast) be attributed to the delays?

    • nuttmedia

      Could be, but looks more like accelerated outflow in 4q12.

    • I don’t think so. The cost of the building will be around $1 billion/yr though I don’t think that will be completely capitalized. Bear in mind that the entire building cost is less than the spending for the tooling ramp up of one iPhone model. Even without a campus Apple spends the equivalent of one Nimitz-class aircraft carrier every six months (which the Navy takes four years to build). It also seems now to be spending more on tooling than Intel (though I have not confirmed it.)

  • obarthelemy

    Using CapEx as a predictor of production

    1- Assumes sales are capacity-constrained (or that production won’t be turned down if stocks increase ?)
    2- assumes CapEx fixes that (ie, not a sourced part shortage, not a production glitch…)
    3- assumes CapEx is allocated to the right products

  • Sacto_Joe

    Sticking my oar into the water:

    As I said on 2.0, I really think the question needs to devolve to form-fitting the huge Capex increase in ’12, and a bit in ’13, to “what changed in the manufacturing process”. Then we might see a more accurate relationship to device sales.

    To me, the thing that makes the most sense is that cost/complexity of manufacturing went up. I think you’d therefore expect to see that impact in the FOLLOWING years, not in the actual year of sales. So to me, it would make more sense to shift the whole Capex to the right by a year. That shows us that fiscal year 2012 had an abnormally LOW Capex to deal with and thus we could expect much higher volume production and sales. Conversely, fiscal year 2013 would show an abnormally HIGH Capex, correlating with lower volume production and sales growth – which is exactly what happened.

    • handleym

      As far as long delays go, there is also the rumored partnership with TSMC.
      What we know is that
      – the A7 is fabbed by Samsung on 28nm
      – TSMC recently announced that they have 20nm running and producing some (unstated) commercial devices
      – they have 16nm with FinFET on schedule, possibly shipping commercially in late 2014 but most likely early 2015

      If Apple have extended their model of bankrolling manufacturing equipment which is then bespoke to their products for at least some period of time, one can imagine that somehow their money is wrapped up in TSMC. I imagine that TSMC would be loath to make a commitment like “we will ONLY allow Apple to use our 20nm for the next three years”, but I think they’d be happy (for the right price) to commit a fab purely to Apple, or perhaps to commit some less sexy and less public technology (eg some form of 3D packaging) purely to Apple use for three years.

      • charly

        Your suggestion is a bad business for a pure foundry like TSMC

  • Andrew Fields

    What if Apple gets into the “computing apparel” business in 2014?

  • Chaka10

    I think it may be helpful to consider the underlying mechanisms that might drive the apparent correlation of unit sales to CapEx for “product tooling and manufacturing process equipment”. As Apple discloses in its 10-K:

    “Depreciation is computed by use of the straight-line method over the estimated useful lives of

    the assets, which … [is] between two to five years for machinery and equipment, including product tooling and manufacturing process equipment.”

    Thus a fraction of each dollar of such CapEx (between 1/2 to 1/5) translates to D&A expense in the following and subsequent years, until it is fully depreciated. That D&A expense is a component of COGS in the product that is produced by the related capital asset (the tool/machine). The above describe linear relationships, i.e., D&A = m*CapEx, where “m” is the inverse of the useful life of the relevant equipment. Similarly, the number of device units corresponding to a given amount of COGS D&A expense can be calculated as U = n*D&A, where “n” is the per unit D&A in COGS Thus, with certain assumptions, an amount of CapEx by Apple should translate in a linear relationship to an amount of D&A (in subsequent years), which should translate in a linear fashion to a number of device units produced.

    The key assumptions, however, in potentially applying the above to estimating Unit production from CapEx are that:

    (a) the weighted average useful life (the denominator for calculating annual D&A) remains constant from year to year; and

    (b) the per unit COGS D&A expense remains steady from year to year.

    Otherwise, if either of the above isn’t true, the relationship of CapEx to D&A and then to units would be linear in any given year, but differently so from year to year.

    We know, however, that Apple’s GMs have compressed recently (i.e., COGS as a percentage of revenues have gone up). While that doesn’t necessarily mean that the per unit COGS D&A has gone up (since we know ASPs have come down), I also don’t know that we can assume they stayed stable.

    In sum, I think the point of the above analysis is mainly twofold: (i) it’s not surprising to me given the underlying mechanics that there is a linear relationship between CapEx and unit production and (ii) that linear relationship may evolve over the years, i.e., still a line but maybe different slopes (looking at Horace’s chart, it seems that a better “fit” may be two different lines, for the first three/four years and another for the last three/four.

    Note, for perspective, total D&A for F13 was $6.8 billion (including D&A outside of COGS). That compares to total COGS of $106.6 billion. I.e., D&A is a very small component of COGS. The point of this is that while the apparent correlation may be obvious to the eye, and the correlation ratios high, nonetheless, forecasting with this sort of correlation analysis can be highly sensitive to small changes in assumptions (e.g., in COGS D&A per unit).

    I still prefer not to get into details, but it was precisely this sort of correlation analysis (trying to tie device unit sales to another known data point — not CapEx or D&A) that prompted me to raise my iPhone unit estimates for the Sept quarter from mid 30-mm (which was my initial estimate) to something in the 40 mms.