Let op: Tweakers stopt per 2023 met Tweakblogs. In dit artikel leggen we uit waarom we hiervoor hebben gekozen.

AMD's low profits / high PE are no concern: A deepdive

By kidde on dinsdag 1 oktober 2019 10:00 - Comments (16)
Categories: AMD, Beleggen, CPU, Dummies, Gratis advies, Intel, PE, Views: 3.564

Voor de Nederlandse Tweakers: Dit artikel is in het Engels geschreven, om het ook beschikbaar te kunnen stellen aan lezers van de AMD_Stock sub-Reddit

Reading time: ~13 minutes

In which a broken bucket, a comparison to AMD's "fabless" peers, and basic "advanced chip production knowledge" will explain the low profit and high PE of AMD stock. Because - contrary to what flocks of parrots believe, this is not a problem at all. We take a deep dive into AMD's Q2 earnings, and we find some very interesting treasures of information; which other media outlets seem to have missed!

The last months, there was a lot to do about the valuation of AMD stocks. Lots of people look at it's low profit. In 2019, for the SAP 500, a fair valuation of a stock is 25 times its earnings. Apply that to AMD's $0,03 earnings per share, and a stock should be worth $0,75. But now it is at $28! So, flocks believe AMD stock is too expensive. I think they're wrong.


I'll explain why, and at the end of the article you will have all the facts to make up your own mind. To all of you who don't work in a factory, the explanation starts with a simple bucket. Then, we will look at actual AMD and NVIDIA earnings. We use what we learned about the bucket, and in the end we shall see, the numbers from AMD / NVIDIA are not that different.

The bucket model
How do you fill a bucket? Well, that's easy. Pour water in it until full. But now, you find out, there's a hole in the bottom. How, or better, in what case can you fill it? I'll pause for a while to let you think.

Okay, so you know the answer right? You will say: "Just make sure, less water leaks out of the bucket, as you pour in!"

Now, become member of "Hydro energy corp.". We have a river, and it has 3 electricity generators in it, which is generating us a profit. Great! But the water in the river could be low, or could be high: Our electricity sales could change to be low or high, and so does our profit. How do we fix this? Of course, we build a dam!

Hydro dam problems and tricks
But now we have a problem: Behind the dam is a lake (reservoir,) and we need to fill it. The dam and lake are like a bucket with a hole, and we can only fill it if less water flows out - and through our electricity generators - than water flows in. This means, less electricity, less profit! Take the Three Gorges dam for example: Filling it took from 2003 to 2010, all while not working 100%. But the best part? After the lake is full, our generators will generate the same amount of electricity and profit before we started building the dam. Later on, I will use AMD's earnings to show where you find this in AMD's 2019 earnings.


While the lake fills up, the output temporarily decreases

But first, let's think What if it is a sunny hot day? Lots of water will evaporate. Yes, that's our money evaporating! And what if we want to generate more profit for a while? Well, we could make two extra holes in the dam and put two extra generators in the holes. That way, for a while, we can run more water through the dam, through our generators, so generate a higher profit for a while. Of course, this is more or less cheating: People who don't know could think we can always run 5 generators. But of course, this only works until the lake is empty. Later on, I will use NVIDIA's earnings to show where you find this in NVIDIA's Q2 earnings.


So, whose profit is slightly drying up here? Read to the end to find out! Hint: It's _not_ NVidia!!!

Earnings for dummies
Because I am not an accountant, I made a very simple model to understand key indicators of quarterly earnings. If you already know this, skip this paragraph to the next paragraph!


Click for bigger image - Simple revenue / cost model
  • You sell products, what clients pay you is "net revenue".
  • But you had costs to make the products, that's "cost of sales". After you paid that, you're left with "gross margin".
  • From gross margin, you have to pay research and development, marketing and administration costs: These are called "Operating expenses". After you paid that, you're left with "operating income".
  • From "operating income", you have to pay or you receive gains and / or interest on investments you made or money you borrowed. After you paid that, you're left with "Income before taxes".
  • From 'income before taxes', you have to pay taxes (doh!). After that, you're left with "Net income".
  • You can divide "net income" by the amount of shares outstanding, that will be the "profit per stock".
  • If you divide the price of the stock by the "profit per stock", you calculated the "price earnings" ratio.
  • As said, for S&P 500 it's 25, while for AMD it's close to 170.

Fabless peer comparison
I put those revenue and cost numbers for the biggest fabless chip companies, Intel and TSMC in a table. Then, I calculated the numbers as percent of the revenue, so we can compare companies. Please click for a bigger image.


Click for bigger image: AMD's costs as percentage of it's revenues, compared against its peers

If you look at it, it is quite easy to see where AMD's low profit is coming from! Their cost of sales is 59% of their income, much higher than the "cost of sales" of their peers. Looking at what actually is 'cost of sales', it's easy to identify it probably has something to do with inventory. Avid readers will already know where this story is going next, but let's make a few remarks about the peer comparison first, to see why some numbers are different:
  • TSMC is a fab with no design department,
  • Fabless companies like AMD are design departments without a fab,
  • Intel is both; it both has a fab and a design department. So you expect their numbers to be between the fabless companies and TSMC's.
  • Qualcomm has high earnings compared to their cost of sales, mainly because selling patents earns roughly two thirds of their income. Selling patent licenses doesn't involve 'large inventories' , so that's why their "cost of sales" is quite low. It's not the best company to benchmark AMD against.
  • Marvell was unprofitable. Also, not the best company to benchmark AMD against.
  • Xilinx sells FPGA's, these are no consumer parts. They're only bough by companies / government bodies with big wallets. Not the best company to benchmark AMD against.
So, what are the best peers to compare AMD against? Well, first, there's Broadcom: Also a fabless TSMC-client. Then, even better, there's NVIDIA: Not only are they fabless and a client of TSMC, they also sell GPU's, just like AMD! However, remember their CPU sales, if still existing at all, is really tiny compared to AMD.

AMD's hydro dam lake in its earnings, and NVIDIA's 'sky-high' pile of raw wafers
So, the content of the hydro dam can be found in the Q-10 filings, just look for an 'inventory' table. What do we see?

AMD's inventory in H1 2019

Well, actually AMD's "work in process" (WIP) doubled in the first half of 2019! At the same time, AMD and NVIDIA also decreased "Finished goods". This is a good thing: It means, they probably sold "older" products with a rebate. If you keep those "inventory" to long, it will diminish value because the products are getting old before they are sold. That would be like water / profits evaporating from the lake. "Selling with a rebate" to diminish inventory could be seen as "putting more generators in the dam": It will generate more profit for a some time, while draining the part of the lake.

Also interesting is NVIDIA actually halved it's 'Raw material', which probably means they didn't order new "raw wafers" for some time, because they had enough of them in stock. Now as to why they probably had multiple months of raw wafers in stock, I don't know. Maybe there was a shortage some months ago, and they were able to secure production if they ordered a huge pile?

NVIDIA's work in process actually declined.

How AMD added 1 to 3 lakes and made 1 much bigger
So, what caused AMD to have to fill this "work in process" lake?

First, there was this lake of Zen 1. It was "diffused" in the USA, in the lake of GlobalFoundries' 14nm production line. Then it went through the second "Made in..." lake at the "assembly and testing" company, and life was good.

AMD Ryzen 1200: Went through a lake in USA, and one like in China (Red highlight)

Then, 7nm came around. They opted for the 'CPU chiplets' to be 'diffused' by TSMC in Taiwan. But 7nm (double patterning) has much more processing steps then 14nm*! Which means, a bigger lake to fill. However, note that the IO-chips are still "diffused by" GlobalFoundries in the USA. Which means 2 "diffused in" lakes while there was previously one, with one of the two much bigger then before.

*From what I could find, Samsung 14nm (Zen 1) uses "Litho-Etch-Litho-Etch steps, while TSMC 7nm uses Self Aligned Double / Quadruple Patterning (SADP / SAQP). SADP takes more steps, and is also more limited to design for, than LELE.


AMD Ryzen 3900: Collection of two lakes, one in USA and one in Taiwan, then through a third lake in China

Next, for Ryzen 3000 series, AMD uses 3 assembly and test (OSAT) companies*. That's 3 lakes! Because those OSATs will first have to wait until both the parts from the United States (IO chips from GlobalFoundries) and Taiwan are through customs and in, and only then can they start assembling.

Good thing is that one of the three OSATS is actually TSMC, that will save some effort. The other ones are SPIL and TMFE. The Malaysian part of TMFE used to be part of AMD, but they also have facilities in mainland China. SPIL also has a factory in Hsinchu, just like TSMC, so that's convenient. But still, it's a lot of hassle to transport all those delicate wafers.

Now, NVIDIA GeForce 20 series, is 'still' on TSMC 12FFN. Much less steps than AMD Radeon 5700-serie, which is already on 7nm. So, remember: NVIDIA still has to fill this disadvantageous lake, when they switch to 7nm!

Now, AMD's PE is ~170 and deemed far too high, it also causes peoples belief the stock is overvalued. Nvidia's stock has a PE of 40, and lots of people think it will grow, because, err.... Well, because AI and Bitcoin revival, yay!!! NVIDIA is the household name when it comes to AI and Bitcoin, no question.

But what if the supply situation en Q2 between AMD and NVIDIA would have been equal? If AMD didn't have to fill 5 lakes with new parts? Well, if WIP would not have increased, AMD's cost of sales would be less, see the leftmost column in the "@ISO WIP"-scenario below. Profit would have gone up from $35M to $195M. Big difference! If their WIP actually declined by the same amount as NVIDIA's (@NVDA_WIP scenario) profit would have gone up to $220. Even more difference. WIP-stabilization would be a likely scenario for AMD in Q3 / Q4. Which means, as soon as the 5 lakes are filled, the now under-utilized profit generators will all be 100% running again! Great!

How would life have been, if AMD didn't have to build inventory, had abundant raw wafers like NVIDIA, spent less money on marketing Zen 2 / Radeon 5700, end spent less on R&D?

Even more unrealistic scenarios
What if AMD had also stockpiled raw wafers, like NVIDIA did, and kind of "used extra holes in the dam" to drain the lake quicker? Of course, this scenario "@NVDA_WIP_RAW" is not realistic at all, but is just to show how equal AMD is (or is not) to NVIDIA in Q2 '19. Well, AMD's profit would have been $247.

Also, AMD has more "operational costs" (37% of revenue) than NVIDIA (26%), because they both have to design a CPU and a GPU, while NVIDIA only designs a GPU. What if AMD spent only 26% at operational expenses? That @NVDA_WIP_RAW_OP scenario is highly unlikely given AMD has to do CPU's as well, but as revenue increases when the full lineup of Ryzen is selling, operating expenses as a percentage should come down a bit; maybe not 10% though. What could help the 37% operational expense, is if the marketing campaign for the new AMD 7nm portfolio is done. AMD is becoming more of a 'trustworthy' name for cloud providers and enterprise again, so this should happen to some degree. What could increase the marketing costs again, is any aggressive campaign from the side of Intel.

It seems if AMD actually spent "NVIDIA's operating expense percentage", AMD's profit would have gone up to $438M, which is a net profit margin of 28,6% of revenue, with all else equal. Interestingly, in this case, "cost of goods" as percentage of revenue would have diminished from actual 59% to 44! Hey wait, 44%, were have we seen that number...

It's what Avago (Broadcom) actually spends on 'cost of sales'! Do you remember, when we said above that both NVidia and Broadcom were nice companies to benchmark AMD against?

Under NVIDIA's WIP, raw material and operational expense scenarios, AMD's per share profit would have been ~$0.38, and it's PE would have been lowered to ~78.

This is unlikely, it's more likely NVIDIA will temporarily move to AMD:
  • NVIDIA's WIP will increase if they adopt 7nm,
  • NVIDIA will have to buy new raw material sooner or later,
  • With the new competition from AMD, NVIDIA probably needs to spend more on marketing and R&D.
That could mean three things:
1) NVIDIA's net revenue will increase by a lot,
2) NVDA stock goes lower,
3) NVIDIA stock P/E will increase towards AMD's.

Could also be a little of all three.
So, as promised in the beginning: AMD and NVIDIA, without artificial changes and on equal footing, wouldn't be that far apart!

Bonus: What about Intel?
Good news is, Intel's WIP doesn't increase a lot. Watching "Finished goods" in Intel's 10-Q filings actually showed a surprise however: It increased 1H '19! That's quite bad news, these are lots of Intel products which can't compete with AMD anymore, because of AMD's 7nm portfolio introduction. This looks a lot like Intel's (Coffee?)-lake "evaporating by some serious amount".

Alternative / lazy valuation?
Sometimes, one might be tempted to just look at market share and market cap and do a lazy valuation:
  • Intel's market cap is $230B @ P/E 11
  • NVIDIA's market cap is 105B @ P/E 39
Let's take a conservative P/E of 20, a bit below the S&P 500 average. Let's also be conservative, and say AMD is going to take 25% market share of desktop CPU's, laptop CPU's, server CPU's and GPU's before the end of 2020. And let's say AMD market cap would be in relation to it's growing profits.

Then, it's easy to calculate the "lazy evaluation market cap" end of 2020: It should be ~120B, while still being a bit conservative to market share and P/E, but optimistic to AMD's profitability. Currently however, AMD's market cap is 32B. How anyone could say AMD stock is overpriced, I fail to understand.

AMD stock critics think AMD stock is not worth growing above $30, because AMD's low profit. Their stance usually is "Show me the money!" Well, I showed you; it was there all the time in the WIP, right in front of their eyes to see. But they only looked at profit.

What does that mean for the near future?

AMD spent Q1 and Q2 '19 filling the entire supply line; and this line became longer while they were shifting from 14nm to 7nm. This caused their Q2 '19 profit to be "artificially" low. When the supply line is "full", WIP should stop increasing and profit should climb. Also, if revenue goes up because more adoption of AMD's 7nm portfolio, operating expenses as a percentage should come down. That means, starting from Q3, 'net profit' of AMD should start to be more in line with its peers NVIDIA and Broadcom.

Disclaimer: The author is long AMD, XLNX & TSMC, while being short INTC.