By Suhail Capital :
Any sufficiently advanced technology is indistinguishable from magic
Last week's Bloomberg Article on George Hotz stirred up some serious controversy in the self-driving vehicle space. While the main controversy centered on Tesla ( TSLA ) and Mobileye's ( MBLY ) supplier relationship with respect to the former's Autopilot system, we believe this incident has opened the door to a much more interesting debate about where this whole space is heading. As we have not commented on Mobileye since our initial report , we thought this would be a good time to update our views on our short thesis and address recent developments in the space.
So what's all the ruckus about?
Well, as far as the market is concerned, the main controversy centered around a Bloomberg article/interview in which famed hacker George Hotz shared his own self-driving hack as well as disclosed that he had been approached by Elon Musk to work on displacing Mobileye's technology in-house at Tesla. This was followed by a tweet from Citron Research highlighting Mobileye as their short of the year for 2016. The combination of the two led to a prompt drop in Mobileye's stock price. The next day Tesla responded with your expected damage control 'correction' corporate statement that took some shots at what Hotz has accomplished, and defended their supplier relationship. This was naturally complemented by several immediate sell-side analyst defenses of Mobileye, and a brief but limited bounce in the share price. Pretty much your typical growth stock controversy market scenario as far as relevant stock moving news goes. We went from a hacker in a garage is going to displace Mobileye at Tesla and maybe bring down their whole business on Wednesday, to what he accomplished can be done by any one of Mobileye's CTO Professor Amnon Shashua's computer science students and that Tesla still loves Mobileye by Thursday. That market pendulum sure loves to swing!
Now strip emotions out and let's just focus on the facts here.
- Hotz hacked an Acura in his garage and has built some form of a functioning autopilot that is highly AI direct perception driven in its approach, and he took direct shots at Mobileye's technology calling it a 'joke' and 'outdated'
- Elon Musk was impressed enough with what he's doing to offer him a job with the aim of discontinuing Mobileye's technology at Tesla as his primary goal.
That's it as far as we are concerned.
Well, with respect to point #1, it is clear Hotz is not putting Mobileye out of business anytime soon with his hack, and his system is so premature it would be ridiculous for anyone to draw conclusions about Mobileye's future based on what was provided in the interview. All we have here is the impressive work of a clearly bright and highly motivated individual with a track record of 'figuring things out'. There is one big caveat here, and that's that ever since Mobileye went public, Wall Street has gone out of its way to hype up their technology. There is an aura of mystique around the company that has at times bordered on the ridiculous. The type of ridiculous that leads to sell-side models in which Mobileye's ASPs are double what they are today 25 years from now! Hotz, with his public persona, just put a major dent in that mystique. Which is ironic because there is essentially an entire industry of traditional ADAS vision engineers who can demystify this in a few minutes. Or AI vision focused autonomous driving experts as far away as China who make bold claims like Hotz. This is the problem with being the very public leader in a space, and a hot tech IPO. On one end you have a target on your back, and the other end you have people who are not satisfied just selling your very good story. They need to create a better one. Mobileye clearly has that problem. Their initial business model was about NOT BEING CUTTING EDGE. They took a cheaper, quicker, we will find a way to make it work approach to get the mass market and be the leader. Business model wise how can you fault that? But reading about 500 people in Sri Lanka validating images doesn't sell 100x revenue stocks.
As for point #2, this story obviously would not have gotten the coverage it received if it weren't for the fact that Hotz revealed he got Elon Musk's attention with his work. The Tesla 'correction' which we will get to in a moment does not deny that he was offered a job to work on replacing Mobileye with in-house developed technology. This is again not interesting news because of any remotely material financial impact on Mobileye's immediate business model, but because of the dent it puts into the Wall Street created Mobileye mystique. Think about it, Tesla sells 10-15k cars a quarter, annualized that works out to about 1% of the EyeQ chips Mobileye will ship this year. Financially speaking Mobileye could never do business with Tesla again, and it wouldn't impact their bottom line in the least bit. But then why was Tesla mentioned like 20 times on the last Mobileye conference call? The answer to that question is not complicated; Tesla has the most technological mystique of any publicly traded company in this market. Mobileye by virtue of being a key component supplier in their autopilot system has decided to marry themselves to that mystique. But the problem is that Tesla is Tesla, and they are not in the business of giving anyone else credit for what they are doing. This has been clear throughout the entire relationship as Musk has gone out of his way to emphasize the proprietary nature of his company's self-driving ambitions (and their correction response gets into this as they continued to stress their whole sensor suite and more importantly that Tesla's fleet learning algo improvement has zero to do with Mobileye). Do you think Tesla is going to dedicate the resources to replace Mobileye's tech to save 20-30$ a chip down the road? Not a chance. Thus, to us the real notable news here is that Hotz has confirmed what many of us already suspected, Musk thinks he can do it better. Not exactly a shocker considering how much press Autopilot has gotten and how outspoken he is on autonomous driving ambitions. What is shocking is that he only offered a multi-million dollar bonus to replace the technology of a Tier 2 sensor provider that was as of a few months ago being valued at nearly 65% of Tesla's market capitalization.
Combine this news with Musk's tweets a few weeks ago about software hires for the autopilot team, and you don't exactly come out with a ringing endorsement for Mobileye's current share price. Anyway, that's our direct take on this Tony Stark-esque incident, now onto the more interesting stuff .
Tesla's Autopilot and GeoHot's Hacked Acura Raise Much More Interesting Future Questions About The Viability Of Semi-Autonomous Driving
Tesla's correction of the Bloomberg article emphasized the unique software nature of their autopilot, the multiple sensors involved, and did some PR damage control by touting Mobileye as the best at what they do. This was all pretty much expected stuff considering the headline grabbing nature of what transpired the day before. What did however surprise us was Tesla's decision to criticize Hotz's project for its obvious limitations.
Specifically, they wrotethis:
We think it is extremely unlikely that a single person or even a small company that lacks extensive engineering validation capability will be able to produce an autonomous driving system that can be deployed to production vehicles. It may work as a limited demo on a known stretch of road -- Tesla had such a system two years ago -- but then requires enormous resources to debug over millions of miles of widely differing roads.
This is the true problem of autonomy: getting a machine learning system to be 99% correct is relatively easy, but getting it to be 99.9999% correct, which is where it ultimately needs to be, is vastly more difficult. One can see this with the annual machine vision competitions, where the computer will properly identify something as a dog more than 99% of the time, but might occasionally call it a potted plant. Making such mistakes at 70 mph would be highly problematic.
So, Hotz is essentially taking an approach that involves building and testing a system with simplified code predicated on outright direct perception learning from actual human driving, and Tesla is criticizing him for the potential risks involved in doing such a thing without extensive validation capabilities. Yet, here is Tesla, with its 'extensive engineering validation capabilities', sticking its definitely not 99.9999% reliable yet Autopilot system in the hands of 40,000 eager consumers. What's more concerning Hotz's machine learning system confusing a dog for a potted plant with him behind the wheel building out his capabilities and collecting data, or 1/40,000 amped up Model S driver engaging Tesla Autopilot and failing to react when the car misses a lane marking and decides to accelerate towards oncoming traffic?
We wonder what Google ( GOOG ) ( GOOGL ) thinks about all of this as they have taken the painstakingly slow and safe approach here versus Tesla who brought (albeit quite impressively so far) Autopilot to market in less than two years.
Here is what they had to say about the flaws of semi-autonomy in their mostrecentmonthly self-driving car project update:
People didn't pay attention like they should have. We saw some silly behavior, including someone who turned around and searched the back seat for his laptop to charge his phone -- while travelling 65mph down the freeway! We saw human nature at work: people trust technology very quickly once they see it works. As a result, it's difficult for them to dip in and out of the task of driving when they are encouraged to switch off and relax. We did spend some time thinking about ways we could build features to address what is often referred to as "The Handoff Problem" -- keeping drivers engaged enough that they can take control of driving as needed. The industry knows this is a big challenge, and they're spending lots of time and effort trying to solve this. One study by the Virginia Tech Transportation Institute found that drivers required somewhere between five and eight seconds to safely regain control of a semi-autonomous system. In an NHTSA study published in August 2015, some participants took up to 17 seconds to respond to alerts and take control of the vehicle -- in that time they'd have covered more than a quarter of a mile at highway speeds. There's also the challenge of context -- once you take back control, do you have enough understanding of what's going on around the vehicle to make the right decision? In the end, our tests led us to our decision to develop vehicles that could drive themselves from point A to B, with no human intervention. (We were also persuaded by the opportunity to help everyone get around, not just people who can drive.) Everyone thinks getting a car to drive itself is hard. It is. But we suspect it's probably just as hard to get people to pay attention when they're bored or tired and the technology is saying "don't worry, I've got this...for now."
Kind of hard to disagree with their concerns, and it seems they are not alone. BMW's CEO took a shot at Tesla's autopilot, dubbing it an 'app industry' approach that doesn't comport with the automotive space where you need to launch a car that is 100% reliable versus simply putting out something that can be done but definitely not completely ready to be put in the hands of consumers yet. Ford ( F ) has also decided to skip Level 3 Semi-Autonomy and move right to Level 4, and word is other automakers are starting to consider the same thing. The growing consensus is, generally speaking, focusing on an HMI to address the human problem in level 3 semi-autonomy is simply not worth the risk, and that resources are better spent focusing on Level 4 or in Google's and some other cases complete Level 5 autonomy.
So why bring this up here?
We personally love what Tesla is doing, and for that matter what Hotz is trying to do. Can you imagine Uber (UBER) retrofitting an entire fleet of their cars with his technology? What do you think that would do for advancing AI trained driving vision systems? The problem is we are now much more confident that this experimenting is either going to end or be brought under a very tight regulatory leash. Tesla is already dialing back on the Autopilot by bringing restrictions to its new upgrade which we have no doubt are going to anger a lot of vocal Model S owners who paid $3,000 for what right now can best be described as a unique gadget whose owners love showing off and tinkering with it, and are always one unfortunate catastrophic accident away from a media and industry firestorm. We also imagine it is a matter of time before the inevitable regulatory backlash regarding Hotz's hacking gains steam. How long before some politician starts making a name with the, 'we can't just have 20-somethings building self-driving cars in their garage and then putting them on public highways can we' message? A couple months max? No matter how you cut it, we are quite confident that the self-driving space is approaching a critical turning point. There is going to be a lot more structure, and we are confident the big loser in all of this will be those with near term semi-autonomous ambitions. We also think this changes the game for the technologies involved on the road to full autonomy, and who the winners might end up being there. With less focus on incremental near term semi-autonomous features, we expect a lot more focus to shift towards LIDAR, high precision GPS, and V2V communication. In particular, we expect LIDAR to be a big beneficiary. Prices have already come down drastically with Veldoyne expected to ship sub $500 systems in the first quarter of 2016 and Quanergy launching a $250 solid state system next month. As more automobile companies focus on Level 4 and Level 5 roadmaps, we expect LIDAR and high precision GPS adoption to accelerate. This isn't just speculation on our part; at a recent industry conference, Delphi's CTO stated , "We found as we looked at the data that LIDAR is going to be instrumental, we think, to fully automated future vehicles." Then you have Mercedes which had seemingly been confident that they can do without LIDAR changing their tune by making investments in the space. The much talked about Nissan Piloted Drive relies on a specially developed high spec miniature laser scanner to allow the car to weave through tight spaces.
How Does This All Impact Mobileye?
Well, on one end there is a degree of good news here as automakers will probably focus more on their near term ADAS plans where Mobileye is a dominant and well positioned player. The bad news is that Mobileye's almost religious argument for a mono vision dominated autonomous future is clearly dead on arrival. The related sensor tech is just going to be way too cheap and way too reliable by the time these cars start hitting the road, and thus Mobileye will be a smaller piece of the pie. Not the biggest deal until you look at sell-side models like the recent Evercore report which has Mobileye ASPs rising to over $100 in 2040.
Why Is the Sell-Side So Stubborn On This Name?
Honestly, reading sell-side research on Mobileye makes us cringe. It's as if their goal in life is to drive unwitting investors to lose money on this stock. Below is the change in Mobileye short interest since we published our report in mid-August.
You'd think this would get some of these geniuses on the sell-side to back off a bit, but that has not been the case.
For example, how can you not cringe when you read comments like this:
"Even though machine perception is likely to be commoditized over next few years, each successive moat may be even wider."-Evercore, MBLY DEC 7
So, they have a 4-7 year lead in deep learning, and a potential inestimable lead in artificial super human intelligence down the road. Mobileye hadn't even mentioned the phrase deep learning until Drive PX popped onto the scene, and all of a sudden they are untouchable here? Do you think this analyst spoke to anyone outside of Mobileye's CTO in the AI Driving Vision space because we can tell him there are some people in the field who would die laughing after reading such a statement. Who are we to judge these engineers, as we are not AI experts. All we can say is that Hotz thinks their technology is a 'joke' and he's made AI his life obsession for the past few years. Maybe just maybe there is a middle of the road that doesn't essentially require stating:
Mobileye's main source of revenue for the next three to four years will be commoditized, but they will dominate cutting edge future technology I have little understanding of for the coming twenty years!!
We do often wonder what it is about growth darlings that makes analysts never ever want to back down. Consider the fact that several analysts are willing to concede Mobileye will see serious commoditization pressure in ADAS, and some are even of the view that ADAS adoption rates could disappoint. Yet, they all seem to have these elaborate projections for autonomous/semi-autonomous future domination as well as strong views on said technology today. You think any of these guys realize that in the time since we published on Mobileye our Nvidia (NVDA) long actually outperformed our Mobileye short. Literally, Nvidia shares are up over 50% versus a 35% decline in Mobileye. There is essentially no material news out of Nvidia and yet the world's leading GPU maker rerated up 50% in the span of months. Yet, here they are all still pounding away on MBLY stock, and having to defend it from threats from 26-year-old whiz kid garage hackers. This to us is remarkable, and here is why.
Consider for a second that we are completely wrong with respect to our views on semi-autonomous driving, and Mobileye's future role as a dominant sensor in autonomous driving. We are willing to accept that and despite our rapidly rising conviction here acknowledge a probability of being proven wrong on this topic. Certainty here is extremely difficult to achieve even if you start to see evidence of a strong thesis materializing, yet this doesn't seem to give most sell-siders pause. Meanwhile they seem much less interested in current developments in the space that have a greater bearing on Mobileye's prospects or for that matter doing a better job of explaining what it is the company actually does.
One thing that still bothers us a lot is this general view of Mobileye as some sort ADAS OS that the Tier 1's simply buy and presto achieve functionality once they plug it into a camera. Tesla has gone to great lengths to shed the off-the-shelf product view with respect to their ADAS, but that doesn't mean the others don't also feel the same way. Mobileye provides vision algorithms on a chip, they do not do the complex control system integration. Yet, we can't recall a single report that has gone to great lengths to explain what this means. There is a reason the Tier 1's working with Mobileye still hire teams of vision algorithm engineers and product feature development engineers, and often work with related contract manufacturers to develop their complex ADAS systems. Also all these players have invested heavily in sensor fusion capabilities to develop the related ECUs, and in many cases in related automobile vision technology capabilities that they are doing in-house despite using Mobileye.
To put the recent hoopla around GeoHot's hack in perspective, just consider the below project description from an employee at Mobileye's largest Tier 1 customer TRW for a rear-camera AEB system for a model 2018 year launch.
Technical Specialist, Camera & RadarTRWJune 2014 - January 2015 (8 months)I was the lead embedded algorithm engineer developing an intelligent rear-view camera to detect and classify objects behind a passenger car. This safety sensor activates automatic braking to prevent the vehicle from impacting objects during backup maneuvers. I designed the machine vision algorithms and wrote embedded C code capable of real-time execution (30 frames per second) on a TI-TDA2x processor. I was the technical manager for a team of 4 contract engineers working in India. These engineers worked under my guidance developing other crucial components of this safety sensor. Intended for MY 2018, domestic passenger car production.
Mobileye is supposedly planning on getting into rear AEB and surround view, yet here we have a team of four at Mobileye's largest customer developing the machine vision algorithms in-house on a TDA2X processor. Hmm, not exactly the type of stuff that quarter century monopolies are made of.
What are we supposed to read into TRW's parent ZF Friedrichshafen's acquisition of the vision algorithm engineering software group of HDLE GmbH? These guys are supposedly developing the in-house surround view capabilities at ZF, where does that fit in with Mobileye down the road?
And how about this recent pamphlet from ZF on Automated Driving which highlights stereo technology as well as mono?
Or how about this recent job posting for an Image Processing Algorithm Engineer from another Mobileye Tier 1 partner Magna.
Magna Electronics, an operating unit of Magna International, provides innovative electronic systems through manufacturing facilities and engineering divisions located all over the world. They design and supply the unique electronic components and sub-systems to meet future vehicle system needs.
• Development (design, implement, validate and debug) of state of the art camera based real time image processing and computer vision algorithms for Drive Assistance Systems (such as feature extraction & matching, tracking, object detection, calibration, distortion and perspective correction, zooming, overlays, brightness control, image transformation) from concept phase to production
• Definition, integration, testing, delivery and maintenance of automotive vehicle control software for applications
• Select and design the image processing and enhancement algorithm for the specific feature set.
• Benchmark competitor's solutions and determine best ways to improve or surpass the competition.
• Develop application image processing algorithm software to meet Systems Level Specifications.
• Work closely with systems engineers to develop algorithm requirements and designs.
• Support customer meetings as technical expert for camera-based image processing algorithms.
Are we supposed to conclude they are building their own camera? Can we conclude they have no aspirations at any point down the road to do so?
Our research presently indicates that is not the case, but then again we also don't expect current partners to advertise their future competitive road map plans to their existing suppliers. What we do know with confidence is that the Tier 1's are developing significant in-house vision algorithm development resources by the simple virtue of working with Mobileye. At the worst, this gives us conviction in the view that when combined with related competition this will continually exert downwards pressure on ASPs as ADAS adoption ramps. We've gone into this detail to prove a point about Wall Street analysts reading too much into things that don't matter and simultaneously speculating about future technology that clearly none of them understand.
What we can tell you is there has been a whole bunch of stuff going on in the space worthy of keeping an eye on that gets virtually no sell-side coverage.
Here is some reading for those who actually like to dig around. Note, almost all of this is during the past few weeks. For those looking for more, feel free to reach out to us.
- Hyundai Mobis considers developing chips for autonomous driving
- Recent job posting from Mobis for Stereo Camera and Mono Camera development projects
- Toshiba begins shipping their image processor to Denso
- Toyota invest 1 billion Yen in AI firm
- Blackberry and ADASWORKS announce partnership
- Baidu sets up autonomous car business unit
- Denso i nvests in deep learning image processing company
Some Thoughts On The Stock Side Of Things
This article really is meant to be about shorting Mobileye stock. We think we did a good enough job laying that out the first time around, and that's why we have kept silent on the matter since we first published. But to be clear ,our views on this short haven't changed much since our initial report, and we don't plan on getting into valuation here again. It is obviously less compelling at $40 than it was at $63, but that is not exactly saying much. The stock could fall another 50% from here, and this debate could simply be recycled for whether it's going to be heading to $10 or $30 from there.
The news flow has kind of gone as we expected with the surprises so far being heavily weighted to the negative side. The Hotz/Musk incident definitely took us by surprise as you rarely expect to see such embarrassing headlines enter the public domain on a company like this. Add in our views on semi-autonomous going forward, our expectations for Mobileye ADAS volumes in 2016, and generally the worst tape we have seen for hyped growth tech stocks in five years and we guess you could say we are modestly more negative on the name now. The macro environment for these names has definitely crept up on us, as for the first time in a while there seem to be no fads in publicly listed high beta tech (that is assuming you don't view Amazon, Google, Netflix, Facebook as high beta tech). With that in mind we have a hard time seeing how the stock doesn't make it into the mid $20s before anyone is going to want to stick their neck out on it again. It's even more of a valuation outlier today, and there are now so many other former high flyers worth looking at for anyone willing to try bottom fishing in growth tech names.
See also Going Shopping: Chicken Vs. Beef on seekingalpha.com
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.