Why Bother?

With any new technology, the first question that comes to my mind is “why bother?”. And the reason I ask that is because usually I fail to see a tangible societal and economical benefit. Just being “cool” doesn’t impress me. But solving problems does. I guess I always have my investor hat on. And I’m an old soul.

Usually, these new technologies win the battle of ideas over me. And usually that’s because many “cool” new technologies have follow-on effects that end up (after 3,4,5 iterations) with those tangible societal and economic benefits that impress me. With Autonomous Vehicles (AVs), I didn’t have that problem. 

In NXP Semiconductor’s Investor Day a few months ago, the head of their Autos division had a simple slide which boiled it all down to 3 indisputable desirable outcomes in transportation. He called it the 3 Zeros:

1. Zero Accidents
2. Zero Congestion
3. Zero Emissions

It’s hard to argue against these outcomes. I doubt we’ll ever get to Zero. But the message is clear: All 3 – Accidents, Congestion and Emissions – are massive societal problems with obvious economic benefits. #1 and #2 are benefits directly associated with AVs. 

Zero #1 is a bigger problem than most people imagine. According to Waymo’s site (that’s Google’s AV company), there were 1.25 million deaths globally due to car crashes in 2014. 94% of these crashes, says Waymo, were undoubtedly the fault of the driver. Speeding, Alcohol, Distraction and Drowsiness are the main causes. We can assume that AVs won’t have the tendency to speed, drink, text or sleep. 

Zero #2 is easier to solve when data and rationality call the shots. We humans are not particularly good at route optimization – it doesn’t come naturally to us. With the growth in AVs, two things can happen: 

1. Better route optimization, because AV’s have no choice but to always follow a navigation system.
2. More ride-hailing services and lesser car-ownership – this means fewer cars per 100 people. Less Traffic.

Zero #3 isn’t directly related to AVs but it is related to Electric Vehicles (EVs). But #3 and, consequently, EVs bolster the case for digging into the Automotive sector even more. This sector sits right at the starting point of two huge disruptions: Autonomy and Electrification. They look like independent concepts but there are common threads, the most important of which is Data. 

From Evolution to Disruption

Cars used to be just mechanical things. Over the last 20 years, it went through a period of increased electrification – not of the drivetrain but of smaller components like the infotainment system. Much of the evolution has happened in the last 10 years, as the advent of the smartphone gave car companies ideas. I need to emphasize two key words in this last sentence: Evolution and Smartphones. 

Since the Model T sped off Ford’s assembly lines in 1908, cars evolved from purely mechanical machines to slightly smarter mechanical machines. The evolution was gradual, only speeding up a bit in the last 10 years. Since then a lot has changed. Just in the last 3 years, scientists and engineers have become confident about Driverless Cars. And this happened around the same time that Electric Cars went mainstream, thanks to Tesla. This is no longer a slow evolution. This is a disruption. And it’s unfolding right now. 

Let’s take a step back and just think how the stars have aligned for Autonomous & Electric Vehicles: 

1. Cloud Computing went mainstream at the Enterprise Level 2-3 years ago. 
2. That unleashed Artificial Intelligence algorithms across many work processes. 
3. 5G is finally becoming a reality.
4. Tesla has proved that Electric Cars can be sexy. 
5. A lot of time and money are being spent on moving beyond Lithium Ion batteries.

All put together, this is a disruption. No doubt. This changes everything in a car – from the way it operates to its drivetrains. At some level a car will still be about kinetic motion. But the way these stars have aligned, in 3-5 years a car will be more of a computer than a locomotive. It’ll be a computer on wheels. 

Level Up

While the jump from the good ole Internal Combustion engine to Computers on Wheels is quite the disruption, there is a lot of work to be done. I don’t want to dismiss the evolution of the internal combustion engine as a trivial thing. Nowadays, many cars already have millions of lines of code running through their veins. But that just sets us up for the disruption, which will probably be more of a step-function rather than one quantum leap forward. 

There are approximately 5 steps in this step-function. There’s an organization called the SAE, which has provided Auto companies with a framework with specific goal posts. You can download the document here. The two key variables, as you can imagine, are: 

1. Amount of Driver involvement
2. Amount of Automation

Here’s what I understood from that document, with percentage breakdowns of each variable. The breakdown is entirely my estimate (not SAE’s):

Level 1: Most functions need a human being. One of the following functions are automated: acceleration, steering or braking. By automated, I really mean assisted. Driver Involvement: 90%, Automation: 10%. 

Level 2: At least 2 of the functions mentioned above are assisted, with a little more automation. So, imagine “cruise control”, but with the foot off the pedal and hands off the steering wheel, on a highway. Still, a driver needs to be behind the wheel in case the “environment” changes. Driver Involvement: 70%, Automation: 30%. 

Level 3: All 4 of the functions mentioned above are automated. Drivers can relax for most of the time but not always. I was trying to decipher from SAE’s document and here’s the best translation I could surmise: Simple, pre-determined routes need no driver involvement. But something like city traffic does. So, a bus or a truck route between 2 cities along a highway can be fully automated, but a driver will need to get involved the moment the truck enters city limits. The big difference between Level 2 and Level 3 is “safety protocols”. Basically, if a child comes running in front of the truck, the truck should brake suddenly. But it should also be able to reliably distinguish between a child and a beer can on the street. Driver Involvement: 30%, Automation: 70%. 

Level 4: Now we’re in “Fully Autonomous” territory. All functions, safety protocols, and ad-hoc decisions are carried about without human involvement. But this level of autonomy may be “domain-specific”. So, commercial vehicles may have a different system than cars used in ride-hailing services. Driver Involvement: 10%, Automation: 90%.

Level 5: This is human-level intelligence when it comes to driving; without the tendency to speed, drink, text or sleep on the wheel. Driver Involvement: 0%, Automation: 100%. This is the holy-grail in the AV world. We take driving for granted, but it is an incredibly complex process for a computer. Go ahead, pat yourself on the back. 

Today, many companies have achieved Level 2 and one or two have even gone on to Level 3. From what I’ve been gathering, it seems that the jump from Level 3 to Level 4 will be the hardest. We may be 3-5 years away from that.

The easiest use-cases that can materialize in the next 1-2 years would be Warehouse Forklifts (Amazon already has these), Construction Equipment, Buses within a ring-fenced compound, and Delivery Drones. All of these can be categorized as Level 3. Given a very specific route, they can already be trained to be Autonomous. One of the most underappreciated aspects about Buylyst holdings Caterpillar and Volvo is that they’re investing heavily in Autonomous and Electric Construction Equipment. To me, Level 3 vehicles in those sectors is a no-brainer. Of course, the vehicle will also be trained to lift heavy material. But that doesn’t seem to be as complicated as AVs on city streets. 

Level 3 is here. But in small doses, and in small areas. To let it out into the wild needs more work. 

Lifeblood: Data

As the dream team of mechanical, electric, and computer engineers figure out a way to go from Level 2 to 5, the main lifeblood is Data. The steps from Level 2 to 5 is an AI problem. 

Any AI problem has two main steps:

1. Training
2. Inference

AVs must be trained using real data from the streets (literally). And then that data must be used to make real-time decisions. This is what companies like Waymo and Uber are doing. They’re collecting as much data as possible while they build programs and hardware systems to handle and process all that data. Then once a lot of the decisions are hard-coded into the car, they will put it through “out-of-sample” tests to see how it reacts in real-world situations. The last term there – out-of-sample – is something I borrowed from my days in Quantitative Analysis. The idea then was to use past data to predict future patterns. So, you’d build a model based on historical data. And then test how the models held up on a set of historical data that wasn’t part of the “training sample”. In AV’s, this test is much more complicated. There are 2 main reasons for that: 

1. The car needs to be trained to think rather than replicate the same algorithm on different sets of data. It needs to build instructions on the fly. Hence, AI. 
2. The sheer volume of data in Training and in Inference is mind-boggling.

On #1, I’ve explored AI in some detail in Artificial Intelligence is Real, so I won’t repeat it here. But the key message there is: it takes a lot of data – and good quality data – to train these AI programs, which leads us to #2. I tried to find a reliable estimate of just how much data we’re talking about here. But I found varying estimates, which makes sense because we’re talking about 5 levels of Automation. But I thought it’s best to use data from companies who are actually working hard at making the leap from Level 2 to Level 5. I looked to NXP, Micron and Intel. 

Micron looks at this problem from a Memory standpoint. They estimate how much DRAM is needed to carry out the AI processes needed to make the jump. 

While the charts look dramatic, Intel puts it all in context for us. They call it the 4 Terabyte challenge. Intel estimates that each AV will generate about 4 Terabytes per hour by 2020. That’s about the same amount of data generated by 3,000 internet users per day. Earlier, I had mentioned that AVs will be more like computers on wheels. But it looks like they will be more like Datacenters on Wheels. Can you imagine the amount of data generated and stored as millions of Internal Combustion cars make the leap?

Autonomous Value-Chain

The way cars will look inside and out will probably change dramatically as the 2 big disruptions take hold. The biggest difference will probably be inside the car – the chassis, drivetrains, electrical circuitry etc. As cars morph into becoming moving data-centers, there are 6 big changes that I imagine will happen faster than we think. The brunt of the expense will be borne by Auto manufacturers. They’ve already started investing heavily. Many have embraced the disruption. As far as I can see, the metamorphosis consists of these:

Sensing: This is where data is collected real-time. The number of sensors needed to go from Level 1 or 2 to Level 5 will multiply by many factors. An AV needs to be reliably observant about every nanometer of its surroundings. The most common technologies used today are Cameras, LiDar sensors, Radars, other smaller sensors. Data is collected via images, light, sound, and radio signals. As you can imagine, redundancies will need to be built for everything. 

Software Processing: This is the part where the AV makes sense of data. This is the well-trained AI program that we were talking about before. Once it’s on the street, it needs to handle those 4 Terabytes (or 5 or 10) of data every hour and make good decisions based on that real-time data that it gets from those sensors.

Hardware Processors: This is the silicon that makes the Software Processing possible. These are CPUs, GPUs, FPGAs and ASICs that we see in Datacenters these days. As Moore’s law comes to an end, Car companies will need to think beyond CPUs to process this unprecedented volume of data, just like datacenters.

Drivetrain: As Electrification keeps gaining momentum, Auto companies will need to think about this major shift in conjunction with AV technology. The future that Auto companies will need to be prepared for is this: AVs will be EVs; EVs will be AVs. The actual mechanics of the car will need to work in perfect unison with the AI brain that drives the car. 

AV Macro Ecosystem: I envision this is a Metadata system that AVs hook onto to get real-time information about traffic, weather, events etc. It could also be a source of training data – your AV could gain from another AV’s experience, so your AV can be better at inference.

5G: This is a big one. AVs will not be possible on today’s 4G cellular network. I went through this in Investing in 5G – the sheer volume of data and the intolerance for latency in moving around that data means that a 5G network will be necessary.

Who’s in the Race?

At The Buylyst, we do everything with one goal in mind: Is there an investment opportunity? This is no different. I’ve tried to map the Value-Chain components from the previous section to companies that are heavily involved in AVs. 

Sensing: Sensata, TE Connectivity, Bosch. 

Software: Google (Waymo), Aptiv, Apple, Uber, Lyft.

Hardware: Aptiv, NXP, Intel, Micron, Xilinx, Nvidia. 

Drivetrain: All Auto manufacturers, BorgWarner. 

Ecosystem: Uber, Lyft, Auto manufacturers, Google, Interdigital.

5G: See Investing in 5G. 

This list is not meant to be exhaustive by any means. But it does cover most of the who’s who of AVs. I’ll need to rule out some of these names right away: 

Sensata, Apple, Micron, Xilinx, Interdigital and Nvidia have already been covered in The Buylyst. So, The Buylyst is already reasonably exposed to the AV theme. 

Bosch is a German company, so I’ll leave it out for now because it doesn’t have an ADR. Google is still primarily an advertising company with fingers in many pies. Investing in Google is essentially taking a call on future privacy laws. Although they are better positioned there than Facebook. If I have more time, I’ll look at Google. And then we have Uber and Lyft, which are not public companies. 

Among Auto manufacturers, we have already covered Ford, Tesla and Volvo. GM and Daimler are the other big players. But the danger with Auto companies is that they’re the ones going through the transition – from Internal Combustion engines to AV/EVs. They’re the ones being disrupted and they’re trying to manage it as best they can. This requires heavy capital expenditure and a massive reshuffling of their manufacturing processes and supply chains. Ford and Volvo are already in The Buy List – that’s enough exposure to this disruption for one portfolio.

This leaves: Aptiv. Intel, NXP, TE Connectivity and BorgWarner.

So, what should I do?

As is customary at The Buylyst, the remaining investment ideas will go through a “first-date test” before we can zoom into one or two companies. Based on the latest financial results, here’s how they stack up:

The numbers don’t look great. But the choices are clear. Although Aptiv is really the only “pure play” AV company on this list, their numbers don’t great. The only companies with a respectable Return on Equity number are Intel and NXP. Intel recently acquired a company called Mobileye for its software expertise. But that part of the business is still a small part of Intel. Their base business is still PCs, which is not exactly a growth sector. However, their combination of CPUs+Mobileye may prove to be formidable. At this point, NXP looks to be the prime candidate. Autos make up about 50% of its revenues, and the segment is almost entirely dedicated to AVs. 

Time to work.