Crypto Concerns, GPU Sizing, Vega Strategy, And Much Needed Noise Filtering


I’m bullish on AMD shares, but reading all these articles on SA as well as industry consulting reports/sell-side research has clearly demonstrated to me that there is a lot of noise around the name that needs to be cleared up, and a lot of interesting elements that are really not being discussed. This is all very surprising when you consider how much is written on this stock.

Here are 4 key issues I want to focus on in this report:

There is a view on SA that AMD has somehow sandbagged guidance in 2017, and thus that H2 numbers which should include a significant Epyc ramp will crush consensus. In my opinion, this couldn’t be further from the truth. Consensus comfortability with AMD crypto-related risks is a major head-scratcher. They were too bearish on the topic for most of H2 2017 and early 2018 and now seem to have gotten too bullish. The sell-side and industry consultants have proven to be woefully bad at assessing both crypto demand and understanding its interplay within the context of the broader GPU market. AMD’s strategy around Vega is worth a closer look. AMD’s console story doesn’t get the press it deserves. There is a positive upside story here that deserves a lot more attention.


Management’s lack of reporting transparency is partly to blame for a lot of this noise. AMD refuses to breakout GPU revenue. I can only speculate as to why this is, but crypto and their exposure in the high-end probably have a lot to do with it. This is unfortunate because I actually think AMD would benefit from more transparency, as those investors willing to take a longer-term view here are quite aware of the uphill climb they are facing and are in fact in the stock because of the disproportionate upside that said opportunity offers. Furthermore, figuring out the exact size of the GPU business isn’t very difficult. AMD’s financials disclosures since they acquired ATI, combined with management’s occasional desire to call out milestone progress, make this lack of transparency rather pointless. Just take a look..


How Big is AMD’s GPU Business?

To begin the sizing exercise, you need to go all the way back to when AMD acquired ATI. AMD’s 10K disclosures since ATI’s acquisition through 2013 regularly called out the exact annual change in GPU revenue and console royalty revenue that was reported in their graphics segment. And because they reclassified royalty from consumer electronics to graphics in 2008, you have a starting point split to work off of. (note there is a small margin for error here of roughly $30m that was eliminated foundry related revenue in 2009 that AMD didn’t classify which I have estimated on my own, but the margin for error impact is so small, it’s immaterial)


2007

2008

2009

2010

2011

2012

2013

GPU Revenue ($mil)

903

1010

1060

1580

1485

1247

1047

Console Rev ($mil)

89

122

107

83

80

170

1146


(Source AMD 10-Ks 2008-2014)

The most important takeaway from this table is that, since they bought ATI, AMD’s biggest GPU revenue year was 2010. This becomes clearly apparent in 2014 when AMD changed their reporting segment structure. The reclassification of revenue allows you to precisely size the respective segments for 2012/2013. This proves quite handy in the grand scheme of things because AMD’s non-console business then proceeded to free fall in 2014/2015, bottoming out in mid-2016.

AMD Segment

2012


2013

Computing Revenue($mil)

3477

2705

GPU Revenue($mil)

1247

1047

Console Revenue($mil)

170

1146

Enterprise/Embed/Other ($mil)

528

401

(Source: AMD Fiscal 2014 10-K)

AMD Segments

2014

2015

2016


Computing/Graphics Rev($mil)

3132

1805

1967

Enterprise/Embed/Semi-Custom Rev ($mil)

2374

2186

2305

(Source: AMD Fiscal 2016 10-K)

Now, once AMD changed their reporting segments, they ceased calling out annual changes in GPU revenue. This makes the 10-K disclosures rather worthless except for the fact that you can clearly see how much C/G shrunk between 2013 and 2016 largely because of CPU biz and APU mix shift. To finish the process of sizing up GPU, you need to turn to the occasional conference call disclosures AMD has provided since the business started recovering, which ironically are quite handy.


On the Q4 2016 CC, AMD management says this:

“Our client revenue was the highest in seven quarters, and we delivered our highest graphics processor revenue in 11 quarters.”

So, the last time GPU revenue was higher was when they did roughly $300ml in q1 2014. This proves to be a useful starting point for 2017 as they call out GPU for being seasonally down in on the q1 call.

Then, on the Q3 2017 CC, AMD management gives you this little tidbit on GPU:

“In graphics, we achieved record GPU revenue in the quarter based on significantly improved ASPs and higher unit shipments from a year ago.”


Record revenue in GPU is a nice milestone marker because the biggest raw GPU revenue quarter AMD has had since it acquired ATI is roughly $410 million back in 2010. So, AMD crossed this mark for the first time in Q3, which comports nicely with the seasonal reported cadence we have from the end of Q4 2016 through Q3. AMD then goes on to call out another record quarter in Q4 but clearly doesn’t call out a GPU record for the year.

They also provide another nice piece of info in the Q4 2017 CC Q&A:

“Yeah absolutely Mark. So, look on the Computing and Graphic segment, we grew about $140 million sequentially. And if I look at that growth, it was across Ryzen and Radeon. If you look at block chain in particular, our estimates are that it was about a third of the growth, a third of the $140 million. And then the rest of the two thirds are around the GPUs, the other segments of GPUs and Ryzen.”


With Vega doubling Q/Q from the mid Q3 launch, and Ryzen being called out as sequential double-digit grower, you can conclude GPU grew in $50-90ml range sequentially.

Put this altogether and you can accurately conclude AMD’s 2017 GPU revenue was somewhere between $1.45bl and the 2010 all-time high of $1.58bl. Let’s call it $1.5bl.

AMD has also told you they have quantified crypto at somewhere between 5% and 10% of total 2017 revenue. At the mid-point, that works out to $400ml in crypto revenue in 2017 as per management disclosure. Now, let’s draw some very clear noise reducing conclusions.


AMD standalone GPU revenue was clearly 28% of total 2017 revenue and 50% of computing/graphics revenue. So, no more of this 50-70% nonsense I have occasionally come across. AMD ex-crypto grew revenue approximately 15% in 2017. This is almost exactly in line with the guidance they provided at the start of the year, before the mining craze came to town. If you think crypto is slightly bigger than they said, well then, the rest of the business may have slightly underperformed their expectations. They sandbagged nothing in 2017. So, these folks thinking ‘AMD will ‘conservatively grow 40%’ in 2018 need to come back to earth. AMD flat H2 guidance with Epyc ramping says a lot about the rest of the biz. They are clearly lapping Ryzen’s launch, expecting a mildly softer GPU market, and based on what Sony (NYSE:SNE) recently said probably factoring in console momentum to take a bit of a hit by the end of the year (they could prove to be too conservative here). If they have been too bullish on blockchain or Epyc ramps slower than expected, H2 looks like it has downside risk to current consensus.


A Closer Look at the Crypto Impact on AMD’s Business

On the surface, AMD’s crypto quantified exposure works out to 27% of their total GPU biz, but the indirect way crypto impacted their business in 2017 is probably far more significant.

Consider the following:

– AMD was essentially non-existent as a NVIDIA (NASDAQ:NVDA) competitor in the high-end gaming, workstation, and datacenter markets in 2017.

– Pre-crypto mania AMD’s mainstream PC gaming value proposition with Radeon 400/500 series vs NVIDIA was also not compelling.


– GlobalFoundries is more capacity-constrained, and, with semi-custom/server/Vega launch/Zen launch, AMD likely had very little production elasticity with respect to crypto demand.

– AMD’s exposure to the lower end of the market outside pure channel AIB discrete desktop is significant.

Taking all these factors into account, one can conclude AMD’s Polaris cards were a huge beneficiary of alt-coin mania in 2017, and that demand here has almost been exclusively crypto-driven. So, when looking at the AIB channel business the crypto exposure is no joke. The good news is it seems AMD knows this and was happy to take the margin boost in 2017 as an opportunity to buy time ironing out the kinks in Vega. The bad news is it seems AMD basically caught a very sizeable broad GPU biz ASP uplift in 2017 that had nothing to do with the value proposition of their chips. When this reverses course, it could be a visible margin headache for AMD.


Why will it reverse course you may ask?

I don’t want to get into a detailed crypto analysis here, but there are some very glaring issues with sell-side and mainstream takes on the interplay.

The most obvious is unit demand over the last 12 months. The number most commonly cited by the press is 3 million cards in 2017. To be clear, a simple understanding of how alt-coin GPU mining works would show this number is way off, and you can go one step further and confirm this by talking to large GPU mining farms. If you think miners drove less than 13 million in GPU unit demand over the last 12 months, you need to do better research. The real mystery unit wise is how much more than 13 million. A little less obvious is the ROI dynamics and interplay with manufacturers. I deal with this in more detail in a marketplace piece, but in a nutshell, these two are deeply connected at the large farm level. ROI expectations are very high, which means short of a significant boost in GPU compute at same value, even a moderately stable crypto price environment is bearish. Following on point two, it seems very few people seem to realize that you are dealing with a commodity bust here. I can respect the failure of technologists to see it this way, but sell-side analysts have no excuse. Yet, they all have focused on coin price swings expecting some sort of more related correlation there instead of tracking hash rates and large mining farm ROI metrics. The crypto bust should play out in the same manner as a real estate bust, physical mining bust, or the recent fracking bust. The supply-chain impact always has a decent lag to the initial price swoon in whatever underlying commodity has been speculated upon. In this case, we are talking about GPUs, DRAM, motherboards, and power modules just to name a few of the component suppliers that are going to take a hit. And you will know when that bust has arrived not by what Bitcoin and Ethereum are doing, but by the fact that 99% of the other 1,500+ coins out there are suddenly utterly worthless. I will also add there is a very clear lack of understanding of just how levered GPU mining is to Ethereum. AMD’s CEO has cited the multitude of coins out there that can be mind as evidence of the current environment being less risky. Her rationale is if one crypto coin falls, well, miners can just shift to another more rewarding coin. The problem with this take is she assumes there is some sort of parity across alt-coin land as far dollar value of block rewards out there that GPUs can be used to chase. This is not the case. There is a reason Ethereum is over 70% of the GPU mining market. If Ethereum became unminable tomorrow because of ASICs or proof of stake, that redirected GPU hash power will be left to compete over much smaller block rewards. The fact that AMD’s CEO takes comfort in something that essentially requires a perpetual bubble state in cryptocurrency land is concerning.


Put this all together and I am quite confident we have not come close to scratching the surface on the crypto bubble bust. When coins start shutting down as fast as they were created, I will believe the eye of the storm has passed. The fact that such a high degree of complacency has emerged just at about the time the first signs of the bust part of the investment cycle are emerging in the supply chain is quite concerning.

AMD’S Vega Strategy

While AMD’s ZEN success in CPU has essentially been deemed a homerun, Vega has proved to be a very different story. In one respect, they have accomplished some impressive technological feats with Vega’s next gen compute units, HBM2 memory, infinity fabric, and a high-bandwidth cache controller. Vega 10 is an impressive all-purpose GPU capable of rendering, machine learning, and high-end gaming. That being said it’s not exactly what AMD’s strategy is here, and the recent high-level turnover in the GPU biz may support such a conclusion.


There is no denying that AMD is resource constrained compared to its juggernaut competition in Intel (NASDAQ:INTC) and NVIDIA. What AMD has pulled off in the last few two years with Zen architecture and Ryzen CPUs, their APUs, Epyc servers, and console 4k gaming is nothing short of miraculous. Thus, one can understand the approach they have taken with Vega, and the fact that they are limited with what they can bring to market. This is not NVIDIA with several base die designs and three different fat margin GPU segments complemented by extensive software development resources and marketing muscle. AMD has to pick and choose their battles as they can’t afford any costly strategic blunders. Now, knowing all of this, one is left to explain exactly what AMD was thinking with their Vega gaming card launch strategy.


Vega showed up on the scene late last summer, and very little around this launch went smoothly. These cards were positioned to compete with NVIDIA’s 1070/1080 Geoforce cards. A simple cursory examination of the manufacturing costs involved with these cards would raise immediate questions regarding this business strategy. Between the HBM2 memory, interposers, and specialized packaging required from Advanced Semiconductor Engineering or SPIL these cards had at least $200 in additional costs that needed to be passed on to the customer. This was on top of the fact that Vega is a very large 484mm^2 GPU vs. NVIDIA’s 1070/1080 GP 104 at 314mm^2. Thus, one had to wonder how AMD expected to make any meaningful margin on these pricing them at $399, $499, and $599. And this is just the cost side of the story. On the performance side, AMD’s Vega 10 gaming cards deliver lower frame rates in most mainstream games than their NVIDIA GP104 competition while consuming 40% more power.


So, what’s the value proposition here?

The quick answer is zero, and this is why the likes of Tweaktown have dubbed Vega a “disaster.”

But isn’t that a bit harsh?

I think so!

Vega’s HBMM controller and Rapid-Fire Math features introduce some new elements that when leveraged by game developers (Wolfenstein II is a good example) deliver some impressive performance. Thus, I wouldn’t call Vega a failure. In fact, I think AMD had zero expectations with respect to Vega Radeon 64/56 being competitive gaming cards. Their strategy here is a bit more complex.


Here is my take on what’s going on.

Radeon RX Vega 64/56 were made almost exclusively with the intention of satisfying Apple’s (NASDAQ:AAPL) iMac Pro. Vega’s compute-heavy jack of all trades GPU is exactly what Apple was looking for. Apple wants a piece of the PC Gaming business, and this was an easy cost-effective way for them to get involved. The RX Vega cards give them a GPU that can offer both high-end professional graphics and gaming to their customers at prices NVIDIA will never match. You can also conclude Intel/AMD’s iCore CPU Radeon Vega M GPU partnership is all Apple-driven. Apple’s MacBook Pros currently use Intel Iris Pro graphics, and it appears Apple has decided to go with AMD Vega going forward. This is the only logical reason for this partnership. So, leveraging Intel’s EMIB interconnect bridge technology, Apple gets the low power ultra-thin notebook form factor with AMD gaming graphics tech at their usually demanded super competitive prices. This Apple driver also probably played a decent part in Intel doing whatever it needed to poach Raja from AMD. One could argue that Apple has in fact quietly decided to take a very pro-AMD strategic position as they are keeping a close eye on NVIDIA related threats in other areas they are focusing on.


What’s AMD get out of all of this?

Well, exactly what they want, a high-volume customer that is essentially subsidizing their development roadmap as they launch their assault on the 70% margin dollar share high-end GPU segment markets. Of course, you can’t publicly come out and declare this, and that’s why Vega Radeon PC Gaming has looked so bungled. Every step AMD has made with Vega makes a lot more sense once you consider they are now far more focused on grabbing datacenter and professional market share. The Vega Frontier Edition card clearly was nothing more than a pump primer for developers ahead of their MI25 and Vega 20 GPUs. It would appear AMD is quite content for their Vega 10 cards to go to the type of individuals looking for competitively priced flexible compute GPU for things like off-site rendering, OpenCL testing, gaming development and the likes. You can also conclude crypto mining was a very welcome development for AMD’s Vega gaming launch and GPU development roadmap. Mining demand being blamed for shortages and memory prices took a lot much needed heat off the fact AMD served up something for PC Gamers that frankly they knew wasn’t intended for them. AMD has this luxury in gaming because of how well positioned they are in the console business.


The Death of Gaming Consoles is Greatly Exaggerated

Remember this Wired magazine bold prediction five years ago..

Now consider that Sony sold 19 million PS4s in 2017, and that in April the console achieved the highest units tally since Wii in 2009. While this record-setting April was largely God of War driven, one can’t help but wonder that the impressive console unit numbers over the last year have also had a lot to do with crypto mania. With a solid mainstream PC gaming rig build now costing 50% more than a PS4 or Xbox One, the console value proposition has gotten a big lift. The big beneficiary in all of this is ironically AMD. Console strength this late into a cycle helps them and hurts NVIDIA and may be playing a factor into AMD overall GPU strategy right now. Five years of steadily increasing GPU ASPs culminating in the insanity of crypto mania over the last year clearly benefits AMD’s roadmap. AMD’s console partners are now very well positioned as they start to deliver next-gen semi-custom architecture to them, and game developers are more incentivized to incorporate a lot of what AMD will be bringing to the table graphics tech wise. So, AMD management has a clear incentive at this point to not prioritize the $200-400 segment of the desktop PC gaming market. Focusing on consoles, Apple, notebooks, workstations, professional, and datacenter makes the most sense.


Conclusion

AMD is a phenomenal long-term story at this point, but investors along for the ride should be a little bit more realistic about their expectations. There are some serious near-term risks related to crypto which hopefully now are better quantified. But as I think I have shown in this article, the near-term mining revenue reversal risks of crypto have more than been outweighed by the strategic benefit said mining mania has provided them. AMD’s competition is very strong and not sitting still, but as long as AMD executes their fat margins are there to eat into. Also, we have seen recently in servers and, with Apple, the entire technology ecosystem is literally willing to assist AMD on this journey. If AMD executes everybody in the tech sector but Intel and NVIDIA benefits. How can you not like that set up at his point? It’s an underdog story with a favorite level margin for error which basically means they may not be underdogs at all.

PS- If you enjoyed this research, this is only half the story. NVIDIAs recent earnings report and certain developments have led me to identify some quite intriguing stuff that Wall Street has completely overlooked. The research tackling this and other related ideas can be found on in my SA Marketplace The Razors Edge

Disclosure: I am/we are long AMD.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

Leave a Reply

Your email address will not be published.