Group think is alive and well on Wall Street. At the beginning of 2016, some analysts liked Nvidia's (NASDAQ:NVDA) story, but they didn't like it nearly as much as right now. Every day it seems another sell-side analyst bumps their price targets on the chipmaker leaving only Andrew Left of Citron (who likes the story just not the price) and Well Fargo (the only bank with an underperform rating) to throw shade on shares. After meeting with two industry experts, and exhaustive work on my own, it became clear to me, Nvidia is ahead of just about everyone in the AI and deep learning hardware/software space.
Sure IBM (NYSE:IBM) has Watson beating geniuses at chess and Jeopardy and helping scientists predict global weather trends, but when it comes to real world applications for you and me, Nvidia is carrying the torch. Their position of power didn't just happen, and quite frankly, they may have gotten a bit lucky. I wrote about how autonomous car AI systems work (the kind Nvidia and Tesla (NASDAQ:TSLA) are pursuing in detail in my last article found here).
The fact of the matter is Nvidia's core competency has always been GPUs (graphics processing units), and for the time being at least, these chips are best suited to the challenges and processing needs that AI and deep learning systems require. I started to think, has a chipmaker ever just run away with a whole space?
Market Share of Chipmakers:
In the PC CPU space, there are basically just two chipmakers, Intel (NASDAQ:INTC) and AMD (NYSE:AMD) and things haven't been great for the latter. In the early 2000s, AMD nearly got 50% of the market for a time, now that share sits at about 20%; the market share %s can be seen here:
(Source: CPU Benchmark)
The chart says it all, but anybody who remembers purchasing their last 3 or 4 windows laptops or desktops would have been able to tell you the same. I remember taking some time when deciding on an AMD chip or Intel chip when we bought computers for the firm in 2006. Now unless we are buying a budget desktop box for an intern or contract worker I don't think too hard, I'm buying an Intel powered machine.
Surely chipmakers sought to halt this kind of domination when smartphones came out? If you have read any of my articles, you know how much I think Intel dropped the ball on the smartphone processor space, being effectively completely pushed out. Again a chipmaker would become the super majority producer for smartphone processors; this time Qualcomm (NASDAQ:QCOM) and their Snapdragon series of SoC (system on a chip). I wasn't able to find as good a chart going through the years, but this pie chart which is a snapshot of 1st Half 2016 shows pretty well what's going on in the market share side of things:
Qualcomm would likely have an even higher market share had it not been for Chinese antitrust disputes that gave Media Tek (OTC:MDTKF) an open window, garnering a huge percentage of chips in budget Chinese Androids. Even Samsung (OTC:SSNLF) who makes their own chips realizes it's better to give the customers what they want and provides their smartphones with Qualcomm chips (this is also due to capacity constraints) even though they make their own. The smartphone processor space wasn't as aggressively skewed to one producer than the PC processor, but we can see again one player was able to become dominant.
Looking at GPUs', we see this same dynamic yet again; this next chart looks much the same as PC processors. Looking at the share of discrete desktop GPUs between AMD and NVDA produces this chart:
Here we can see Nvidia is much like the Intel of the GPU space. Again AMD had a shot some time ago, losing yet again to a producer putting out quite frankly superior products (or so my buddies who game tell me). What does all this tell us? Well, it tells us one chipmaker can corner a space quite effectively; Intel did it with PCs, Qualcomm did it with Smartphones and Nvidia themselves did it with GPUs. So is it that easy? Are the other players in the deep learning and AI space likely to be fighting over scraps for the foreseeable future? Before we come to such a conclusion, it makes sense to at least look at the other players' offerings in the space.
Deep Learning and AI Driving Solutions:
Tesla put a giant stamp of approval on Nvidia's PX-2 solution when they announced all Tesla vehicles (including the 3) would be outfitted with the onboard supercomputer from Nvidia going forward. I would say the closest competitor to the PX-2 is probably the NXP/Qualcomm Bluebox. From 35k feet, the PX2 looks like a much more powerful system. The PX2 operates at 8 Teraflops vs. 90k DMIPS on the Bluebox; the PX2 requires 250 watts of power where the Bluebox operates on just 40 watts.
These stats are a bunch of tech jargon that doesn't do much for us as investors or how useful the product is in the real world. Both solutions use sensor fusion effectively using multiple sensors (cameras, radar or LIDAR) to get a real-time 3D picture of the world around it. Both systems use machine learning and AI to help make decisions based on all that information. That's the important part right there and remains the big difference in how companies are tackling the autonomous car space (the part of deep learning and AI I believe will be most visible to people for some time, minus maybe bots).
Companies such as Mobileye's (NYSE:MBLY) current offering at least; programs for every eventuality a car may find itself in and has a specific IFTHEN function where if the car sees X thing it does Y. The deep learning and AI systems of Nvidia and NXP/QCOM learn how to drive much the same way as we do. We may see something that's not exactly like a past experience but similar and thus know what to do, with reasonable doubt. "Learning" from the past are how these AI systems work; they can decipher if something looks like something they saw before and know what they did and with reasonable doubt know what to do again.
They don't need EVERY eventuality programmed into the system because they learn. And because they learn, they will get better over time as they have more data or more "experiences" with which to work, at least that's the idea. Nearly every expert I speak with and industry heavyweights (Elon Musk, George Hotz, Google, et al.) think that AI/Deep learning is the way to go vs. brute programming for autonomous driving.
Mobileye used the latter in its first products but looks to incorporate more deep learning in its EyeQ5 solution, estimated to be available in Q1 2018 at the current time. One year in technology seems like an eternity, and it wouldn't surprise me to see these solutions from Mobileye out sooner than that.
I wondered when AMD would throw its hat in the ring for some time. I wrote before why GPUs are great at dealing with AI/Deep learning, and as we saw earlier, there are two main GPU companies. AMD finally came out with its Radeon Instinct line, which is geared towards deep learning, large data sets, and AI and not AMD's traditional customer, gamers. The AMD Instinct is an open source platform meaning programmers will be able to use it for whatever they like.
I like that they are going open source on this one because it means developers will be able to use the hardware for whatever AI/deep learning tasks they like. Utilizing AMD's chip a programmer could train a computer to trade just like a new trader might learn. Traders often talk about breakouts when price breaks out from a past level that was resistance in the past.
Deep learning would make easy work of a task like identifying potential breakouts and could potentially trade on them as well. With open source, developers will be able to try their hand at deep learning on whatever datasets are relevant to them. AMD is using a familiar form factor (the PCI slot) on the Instinct, effectively opening up AI and Machine Learning to the masses:
The Radeon Instinct cards are expected to ship sometime in the first half of 2017, and I can't wait to see what applications they are used on. The Instinct has 3 different offerings vs. Nvidia with its one Titan X for home AI/Deep Learning tinkerers; the 3 models should bring AI down to a price point for everybody. I can nearly guarantee Instinct cards will be used in an autonomous driving application by somebody though it may be more in a comma.ai type setting where systems are pieced together by new gen hackers, which Hotz clearly is.
The Instinct isn't even on the shelves yet, but at least they are entering the race. No stock listed so far has more to gain from AI/Deep Learning systems taking off and their offering being successful than AMD does. If these systems take off and become the next big thing and AMD gets it right, it could move their needle significantly; a problem you run into when investing in some of the others.
When investing in Qualcomm and thus investing in NXP (NASDAQ:NXPI) to try and gain exposure to trends like AI and deep learning, you run into an oft-encountered issue. The company is so big, has so many different divisions that it becomes increasingly more difficult for something to move the whole company's needle. Qualcomm bought NXP (or is buying as it were) for 38.5Bn vs. a 2015 Revenue of 6.1Bn. Before the merger, NXP expected the company's revenue to grow by 5-7% compound through 2020 and expect much of that growth to come from the automotive space.
Qualcomm has a market cap of nearly 100bn vs. 2015 revenue over 25bn and saw its revenue decline in 2016. It's no wonder why Qualcomm was looking for the next big thing in chips, and NXP was likely a good a bet as any. It needed to find another way to grow. Much of their growth for the next 3 years is expected to come from its NXP acquisition.
If NXP was a standalone (and it still is I guess, though you're capped) and the price was right, it would be a very compelling investment indeed. I'm sure the next iteration of the Bluebox will be even more impressive than the last, but for the time being, when you buy Qualcomm, you know you are buying a company that has a solution on the shelf.
The last chipmaker that has an AI/deep learning solution is Intel with its Knights Corner, Landing and Hill product family that utilizes the Xeon Phi chip. Most of these solutions are currently more applicable to data centers, but most of the GPUs now used in cars were also optimized in the data centers first. I've given Intel a hard time for, oh, about a decade now.
I still can't believe they effectively let smartphone processors pass them by and their stance on embedded graphics certainly cost the company as well. (They thought most all computing could be done with embedded graphics with no need for separate GPUs.) It looks like Intel is taking AI/Deep Learning seriously, but it's still quite tight with any information regarding what exactly the company is putting together.
They have partnered with Mobileye, BMW (OTCPK:BMWYY) and Delphi (NYSE:DLPH). Intel will provide the SoC silicon, Delphi the sensors, BMW the car and Mobileye the software and cameras to make the whole thing purr. On paper, it looks like a great partnership, and it will be exciting to see the final product, but we are still missing a lot of specifics. We know the SoC will be comprised of multiple Xeon cores but whether it replaces or combines with the EyeQ chips is a mystery still (they might not even know yet).
It makes sense for Intel to take over the AI component of the system as Mobileye hasn't been focusing on that aspect but how all systems and responsibilities will be divvied up is still a work in progress. To me Intel has too much legacy baggage; we just learned that PC sales have now shrunk for 5 straight years in a row (source: gartner.com).
Intel would have to execute with this Mobileye, Delphi, and BMW partnership to offset continued PC weakness and we know they won't be a player in the smartphone space, which also is set to shrink. Mobileye and Delphi have perhaps the most solid Rolodex when it comes to relationships with automakers, so if their system becomes a winner, I suppose they could drag Intel along for the ride.
Like most AI, Tensor Chips started in the data center learning tasks by analyzing huge data sets. Could Tensor chips be used in autonomous vehicles? Sure they could, but Google hasn't made clear their intentions on this front if they even have any. Google tends to build hardware when they can't solve a problem with off the rack solutions.
Once hardware manufacturers catch up, they go back to pitting them against one another to provide that hardware at the lowest possible price. See what the company did with Motorola. It showed the smartphone world the direction it wanted to go with Google Now (which is becoming an AI digital assistant) and once that was made clear, it unloaded the hardware side.
Google doesn't want to build cogs or sprockets for the system. No, Google wants to BE the system not just a part of it. Also with Google, you run into the needle moving dilemma on a massive scale. It's hard to move a 550 Billion market cap company, and I'd bet it's not going to be a hardware component that does it. If you are investing in Google to get at autonomous vehicles, you're likely not doing it for the hardware.
Investing in Google for autonomous cars and AI is more a bet that Google will be able to use these tools to bring Google into your everyday even further. I can think of lots of reasons to invest in Google, but the Tensor Chips are not one of them.
How has 2nd and 3rd Place Fared:
We've gone through a lot so far but can come away with two main points. One company can become the super-majority player in the chip space for a specific application. Intel did it; Qualcomm did it, and Nvidia already has done it once. Whether Nvidia can do it again is not so certain and the second thing we have learned is that lots of chipmakers are coming after the AI/deep learning space with solutions of their own. AMD found itself in a tough spot after losing both the CPU and GPU wars to Intel and Nvidia respectively.
Samsung, which is South Korea's biggest company success, was not solely geared to its chipmaking activity still managed to build a nice little business with 17ish% of the smartphone processor space. Media Tek has a 10bn plus market cap with 23% share, but they target lower margin budget phones, so that market cap is not surprising.
You'd expect the top dog to have pricing power and a premium valuation, two attributes that Nvidia currently possesses. There is room for #2 and #3, and they can build nice little businesses in their own right. So who do I think deserves a look for a potential catch-up trade or viable minority player? I think AMD deserves a second look.
AMD was almost as universally hated by analysts for years just as Nvidia is universally loved by analysts today. I think it's clever to go the open source route with their Radeon Instinct chips. Nvidia won the AI/Deep learning space over with the one-two punch of great hardware and solid software. AMD knows they likely can't compete on the software side so what better way to go than open source and great hardware. AMD on a financial basis is coming out of what only can be described as a decade of straight pain.
Nothing went right for the company for a very long time, and all you need to do is look at both its CPU and GPU market share charts to figure out what happened. The company desperately needed a new lease on life, and I think AI/Deep Learning may be just the ticket. We are in the first inning in the AI/Deep Learning ball game; some might even say the game hasn't started yet.
I'm focusing on the AI/Deep Learning side of things for this article. For a great article on other reasons to buy AMD see Austin Craig's recent article here. Do I expect an AMD-powered OEM autonomous car on the road anytime soon? I don't, but I do see a team of engineers and programmers who built an autonomous platform using AMD open source chips scooped up by another company. Hardware has always been the tool, nobody ever asked what kind of computer Facebook (NASDAQ:FB) or Google was built on, they just know it got built on a computer.
Hardware is just the tool that developers use to develop, and that's why I applaud AMD's decision to go open source on Radeon Instinct. AMD has its hat in the ring, and it looks like a good hat. The company has long waited for just one thing to go right and I believe they have finally found that thing. I expect to be hearing about AI/Deep Learning until I'm sick and tired of hearing about it. That's a good thing if you're an investor in these names.
It's still early days in AI/Deep Learning and Autonomous Cars, but Nvidia has clearly pulled out in front. We've seen a chipmaker can dominate an entire industry; it's happened a lot. We've seen there is a host of companies out there trying to stop that from happening. Qualcomm's Snapdragon chips became synonymous with quality, speed, and a cool name; it got people wanting to say they had those chips in their phones. Will Nvidia become so synonymous with quality autonomous cars you'll brag about the chip in your new car as well as the range or level of autonomy? Weirder stuff has happened.
The stock is priced like that might happen. Nobody talked about the chips inside their flip phones, but that changed. When I think of AI/deep learning, I think of GPUs, and when I think of GPUs, I think of Nvidia and AMD/Radeon. Great things are clearly priced into Nvidia. I think if the market prices just good things into AMD, there's significant room for upside.
I'm going to wait and see if AMD tests its 50DMA at 9.55 to stick in my toe. The chart looks like it rolled over and earnings are soon. But I want to be long AMD for at least the next 2-3 innings of the AI/Deep Learning wave and think dips should be bought. I think great things will be done with their chips that could lead to great things in store for the company and hopefully their shareholders, finally.
If you found this article useful and look forward to my next articles regarding AI, Deep Learning, and Autonomous Cars, scroll to the top of the page and hit the follow button. You'll be the first to know when my next article comes out, and it helps me to determine if people are finding my work of value. Thanks again.
Disclosure: I am/we are long NVDA, DLPH.
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.
Additional disclosure: Long AMD Jan 2018 10 Calls
Editor's Note: This article discusses one or more securities that do not trade on a major U.S. exchange. Please be aware of the risks associated with these stocks.