- I have been more bearish than most on Seeking Alpha regarding Intel's tablet plan for quite some time.
- I think 2014 could get continually worse as volumes ramp.
- Impending product releases could offer a slight reprieve.
The last article I penned on Intel (NASDAQ:INTC), simply stated that even the most optimistic views regarding Intel's tablet aspirations would not necessarily translate into substantial growth. The article was published in early January, and since that time, we have had one quarter of earnings and are now able to see some of the impacts of contra revenue.
In this article, I would like to explain why I do not feel the current spending trajectory is sustainable, look at how much the subsidies are costing Intel, and look at the upcoming products to understand why I think the impending release of Sofia and Cherry Trail could offer a little reprieve.
Going For Broke In Tablets
During the most recent quarter, Intel lost ~$930M on ~$150M in revenues (source: INTC.com), and the accounting for contra revenue makes it somewhat difficult to know the exact breakdown of how these losses are occurring.
Shortly after publishing my article, Dr. Stacy Rasgon of Sanford C. Bernstein put out his note on INTC in which he asserts that Intel is spending approximately $51 per chip in contra revenue dollars. Dr. Rasgon's reasoning is very clearly laid out in the article on Barron's for those who are interested.
He also points out that even if you make a few reasonable assumptions that favor Intel, based on the ASP of tablet processors and the volumes Intel could likely compete for, there are limited sockets available to the company.
After the most recent earnings report, contributor Mr. Jeff Groff works out the math, based on CFO comments, for the likely impact of Q2 contra revenue subsidies. The number Mr. Groff comes up with is $178M, which seems to be in stark contrast to Dr. Rasgon's $2B.
Explaining the Differences
Source: Intel full year 2014 outlook guidance from Q4 earnings
Source: Intel Q2 Outlook
CFO Mr. Stacy Smith explains during the Q1 conference call that (my emphasis added):
The big offset to that have been are the dollars that we're providing in terms of contra revenue dollars for tablets. We also talked about it in some depth at the investor meeting. We're not seeing a lot of those in the first quarter, they really build over the course of the year so we have the positives that I outlined and then you will see that offset as a result of you know there is temporary contra revenue dollars that we're providing to our customers as we take our tablets product volume kind of broadly through the different OSs and price points.
Intel is expecting the negative impact from contra revenue dollars to rise in lockstep with tablet volumes as the company presses toward the 40M tablet goal.
So although Mr. Groff's number may seem in stark contrast to the $2B Dr. Rasgon put forth, the math actually works out end to end to ~$2B.
Percentage Gross Margin = 100*[Revenue - (Cost Of Goods Sold)]/Revenue
COS (cost of sales) should be relatively constant, and looking at Q2 guidance reveals a revenue forecast of $13B and GMs of 63%, with a negative .5% impact in tablets. Solving for COS yields $4.81B for Q2, and then using this relatively fixed cost to solve for revenues if we exclude the negative impact from tablets yields $13.178B, hence the implied negative impact of $178M for Q2 - the $178M is the number Jeff comes up with.
Using the same calculation and methodology to work the numbers for full year 2014 yields a COS for full year of $21.2B and an implied revenue, if the negative impact of tablets were excluded, of ~$55B, which would be ~$2.06B above current guidance.
So in actuality, the calculation used by Mr. Groff aligns with Dr. Rasgon's math, and suggests that we could see contra revenue expenditures rise dramatically in the coming quarters.
The Math Isn't Quite This Simple, In My Opinion
Amortizing this ~$2B impact across the 40M units expected to ship in 2014 would average out to $51.60 in contra revenue dollars per tablet chip shipped.
However, this would also imply that Intel had a negative impact of $258M ($51.60 x 5M tablets shipped in Q1) during the most recent quarter.
Intel experienced a 61% YoY decline in mobile revenues. Because contra revenue affects the top line of the mobile operating group segment, and Intel doesn't (as far as I'm aware) share this number explicitly, it makes it fairly difficult to exactly calculate how much contra revenues dollars will negatively impact the company.
What we do know is that the 61% YoY decline equates to a drop of $248M from the same quarter the year prior, and that Intel has been aggressively ramping tablet volumes, meaning revenue without the impact of contra revenue dollars would likely have shown a YoY increase.
Looking at YoY decreases in mobile revenue is an independent data point that is separate from GM, and even looking at contra revenue spending in this fashion suggests a subsidy that is substantially higher than $20 per chip.
Unfortunately, because there isn't a lot of color into the particulars of contra revenue it is hard to perform any precise modeling. The $178M impact in Q2 calculated by Mr. Groff, if using a subsidy of ~$50 per chip, would imply tablet volumes to drop during Q2 to around 3.5M, whereas the $20 stated by Intel would imply tablet volumes to ramp up to ~9M.
Another data point we have is Intel's intended 40M tablet target. Given 5M sales during Q1, the company needs to increase sales during the next three quarters to meet this target.
This is nothing more than a made up scenario that is only being used to demonstrate why I don't believe the math is quite this simple, but 9M tablets in Q2 constitutes 4M tablets more than Q1. A $50 subsidy implies an additional ~$200M in contra revenues above Q1 levels.
Actual revenue will go up, if we assume an ASP of $30 ($30 is just under Intel's list price for these chips, and is about the average of some of the higher end SoCs) by around $120M above the Q1 level.
The end result of this implies that contra revenue goes up by $200M, partially offset by a revenue increase of $120M, with the end figure being Q2 revenues of under $100M and a sequentially higher operating loss.
It's hard to imagine a situation in which volumes are rising substantially but contra revenue subsidies actually outpace revenues enough to cause another 50% sequential decline in mobile revenues. If it turns out this is the case analysts will likely have some difficult questions for the Q2 Q&A.
Q2 will give us a much better indication of how much contra revenue is materially impacting Intel as we'll have a second data point between tablet volumes, YoY and QoQ changes in revenue, and impact to gross margin to get a more accurate sense of the actual impact.
Mobile Is Still Bad Even If We Take Away Contra Revenue
If contra revenue directly detracts from the top line, adding in the ~$250M to assume revenues are flat YoY brings us to $400M. Now if I assume the 5M tablets sold carried an ASP of $30 and represented an accretive volume of chips over last year's figures, meaning a positive additional impact of $150M above Q1's mobile revenues, Q1 revenue would've been in the neighborhood of $556M.
Even fudging the numbers to the positive side by $400M to account for accretive volumes of chips and contra revenue going away wouldn't have made Intel turn a profit in mobile during Q1.
Cherry Trail, Sofia and Broxton
Ashraf Eassa has an extensive write-up on the upcoming mobile products and discusses the three that I would like to point out.
My bearishness on Bay Trail was never one of an absolute performance standpoint, but more from the view that I didn't think the chip would truly achieve leadership, and I didn't think the chip would carry the same margins as Intel's larger Core CPUs, meaning a disproportionately larger number of Bay Trail CPUs would need to be sold to offset declining PC sales.
Intel is claiming these new chips will have much better profitability when compared against Bay Trail.
Source: PCWatch, my editing for grid lines
Above is a Bay Trail die shot, and you can see the GPU comprises ~15-20% or so of the entire core. Using the midpoint of 18% and a die size of ~100 mm^2 for Bay Trail means the GPU is roughly 18 mm^2.
Leaks suggest (as pointed out in Ashraf's article) that Cherry Trail will feature 16 EU graphics cores. Although the GPU cores are tweaked slightly, if we assume they're roughly the same size, this would mean that Cherry Trail's die size, if it were baked on 22nm, would be ~160 mm^2.
Scaling from 22 nm to 14 nm would yield a scaling factor (if we assume perfect scaling) of .4, or a die size of around 65 mm^2. Ashraf guesses that Broxton would be ~60 mm^2, and Cherry Trail would be smaller than Broxton. However, if we look at the changes between Bay Trail and Cherry Trail, and account for scaling, we wind up with Cherry Trail being larger than his estimates for Broxton.
Joel Hruska of ExtremeTech explains Intel's scaling plans for 14nm and beyond compared to other foundries.
Design and layout can focus on different goals, and Intel is pushing more for density at smaller nodes. So it's possible that Cherry Trail will be smaller than what a pure scaling factor would suggest, but on this one we'll likely have to wait and see to get an exact estimate.
As of February, Cherry Trail and Sofia are listed as 2H 2014 products (source: AnandTech). Dual channel memory and 16 EUs will likely put graphics in the neighborhood of Intel's HD4000 line (likely slightly less due to power constraints), but it will be a very sizable jump up from the current offerings in Bay Trail.
Currently Intel is trying to serve the tablet market with Bay Trail. And the chip is as large as competitor chips, but is not a leadership product, and requires the use of subsidies to entice OEMs. However, by trifurcating the chips and offering several products, Intel will be able to attack the different markets with the right size of product to enhance profitability.
Where I typically part with the traditional Intel bull is when it comes down to actually looking at Intel's products and comparing these products against the competition and looking at overall profitability.
Intel is showing an interest in entry, mid and leadership devices. The top of the line devices can afford the higher premium, and if the contra revenue dollars do indeed go away it's likely Intel can start turning a profit on these chips.
In the lower end it's a little harder to call. All these chip vendors are in a multi-dimensional prisoner's dilemma in which they have to maximize profitability for themselves while being competitive from a pricing standpoint for their partners. Regardless of how great one of Intel's products may be, if the company charges too much it opens itself up for being undercut by competitors.
Intel's competitors have tiny chips already on the market for entry-level products. The RockChip RK3188, for example, is 25 mm^2. This is one quarter the size of Bay Trail. (source: Chipworks). Further, we know that Bay Trail is horrendous from a profitability stand point, yet many of these cheaper chips are profitably sold for likely around $15 or so, with some of the higher end chips going for around $30.
In a separate article last year, Ashraf calculated the blended ASP of the larger Core chips to be ~$120. My calculation was a little different from his, and I devised a metric, revenue / mm^2 in order to show the relative profitability of the different products. My article also links to many of the values listed below if you would like more info.
A 300 mm wafer has about 70,000 mm^2 of usable area, and larger die sizes typically have worse yields. Haswell dual core ULV chips are around 170 mm^2, so if we assume 80% yields we would obtain around 330 Haswell ULV chips, which would equate to ~$40k in revenues per wafer, or $.57 / mm^2 for Haswell ULV parts.
The same calculations for Bay Trail, assuming a 100% yield for Bay Trail and an ASP of $37, equals $.36 / mm^2 for Bay Trail, which is just over half of that of the Core CPU line.
Assuming a die size of 60 mm^2 for the follow on products for Bay Trail, 100% yield again and the same ASP of $37, the numbers come out to $.62 / mm^2. An ASP of $30 would work out to $.50 / mm^2.
Note that my calculations, based on yield, are likely overly favorable - we do not know actual die size of shipping products, nor do we know the actual ASP of Intel's tablet chips. $37 would be the very high end, and Intel is not the established leader as they are in the PC space, therefore it may be difficult to command the higher ASPs.
But what these calculations do show is that profitability for Cherry Trail, Broxton and Sofia will likely be much better than Bay Trail if Intel is able to get down to the 60 mm^2 die size for these follow on products.
For the record, I am still quite unsure of Intel's tablet plan. Dr. Rasgon quoted two separate numbers for the BOM subsidies of $20 and $50. Even having to pay an OEM $20 to use a chip that carries a ~$30 ASP cannot be considered a good thing.
But to be honest, I had been planning on running the numbers to share my thoughts on the follow on products, but didn't expect them to come out quite as favorably as they did.
HD4000 wasn't a very potent solution for graphics and Bay Trail used one quarter of the hardware and way less power than the Ivy Bridge graphics chip, so I was also never a believer in the Bay Trail story.
At 14nm, the extra real estate afforded can be used to substantially beef up graphics and should allow for a simultaneous die shrink. 16 EUs used in Cherry Trail, paired with dual channel memory, should offer graphics slightly under HD4000 levels, slightly under because the larger Ivy Bridge Core chips had a much higher power budget. But it will be a massive step up from Bay Trail. Also, Bay Trail's CPU performance was quite good, and the Airmont cores from my understanding are mainly a shrink of the Silvermont cores used in Bay Trail. The extra headroom should allow for slightly higher clocks when compared against Bay Trail, meaning we should see CPU performance increase commensurate with the rise in clock speed.
I think the follow on products will stand a much better chance at being able to be sold profitably than Bay Trail.
But bear in mind I was using what I feel are fairly conservative estimates, so I am not expecting wild profitability in mobile based on the follow on products. I more view it as declining contra revenues and a better product mix allowing the company to stem losses and possibly swing to a slight profit in mobile at some point in 2015 or 2016 - the glass is 1/8 full kind of optimism.
For the balance of 2014, the math leads to some pretty ugly conclusions for Intel. Even Intel's CFO states that the contra revenue impact will build throughout 2014 as volumes ramp so it's entirely possible to see it get worse before it gets better.
Beyond 2015, I still believe upside will be limited as Intel is unlikely to win the iPad or iPhone, and Intel likely won't be able to compete in the cheapest part of the market profitably. Intel is aiming at a moving target and the competition isn't standing still.
And realistically, regardless of how long the benchmark bar graphs are for whatever chip Intel releases, these chips are still destined for mobile where the ASPs and margins are much tighter than the PC space. Compared to the current operating loss, even turning a couple of dollars profit per chip is a much better situation for Intel. An ASP of $30 on operating margins of 20%, which is roughly inline with Intel's blended operating margins, still only represents $6 in profit per chip. Intel plans on selling 40M tablets in 2014, and if the company managed to generate this $6 profit per chip, this would only add $250M in earnings for the full year. The real earnings potential is from stemming the $1B quarterly loss more than actually making money.
And think about this full year figure for a second - Intel lost under $1B in Q1 to mobile. Even if the losses stopped and Intel managed to generate this $250M per year earnings in mobile, it would still take four years to recoup this money. It would take eight years if Intel loses another $1B next quarter. This is overly simplified, but it should drive home the point that this current spending trajectory is highly unsustainable.
Also, I think those who beat the "PC is dying" drum are missing the larger shift, and that's just PC's simply evolving form factors. PCs have been getting smaller since the ENIAC, so this is nothing new. But the difference is the chips are cheaper and there is more competition for these mobile sockets. So the consumer that grabs the Bay Trail tablet instead of the Core i5 laptop is costing Intel some revenue. Intel cannot not make (double negative intended) products for this space, so the company has to release competitive chips in order to remain relevant. In essence, there are some second order effects here that are creating a situation in which Intel's cheaper chips are competing with the more margin accretive larger ones, and it's unavoidable as form factors evolve.
So there are plenty of negatives regarding Intel's mobile situation, but for the optimist, as contra revenue fades and Intel launches smaller products, at least the company should be able to stem mobile losses and stop the bleeding. Even a $200M operating loss in mobile compared to this quarter's $1B loss is better when it comes time to report, but this reprieve isn't likely until we see smaller and more competitive products.
Additional disclosure: I may add to or liquidate my position in INTC at any point.