< Back

“Data is the new oil”

“Data is the new oil”

Why the race for data ownership is the new driving force of the automotive industry.

Yesterday Intel & Mobileye announced that Mobileye will be acquired by Intel for over $15B, an announcement that made waves, given its sheer size and importance, not just for the Israeli tech ecosystem, where Intel is already the largest player, but also worldwide. Much of the discussions yesterday debated the question of whether this deal is good or bad for Israel’s Scale-Up nation, or suggested this is part of the semiconductor industry consolidation.

I would like to suggest a different strategic take on this deal, one that resonates with the words Intel’s CEO, Brian Krzanich, quoted yesterday: “Data is the New Oil”.

My bet is that when we will look at this deal a few years from now, we will choose it as the automotive industry’s watershed moment; The moment we recognized its transition from hardware-centric to data centric; The moment it moved from being about cars with CPUs, to being about computers with wheels. From being focused on the massive, convoluted, hardware components’ logistic ecosystem, to being focused on data & intelligence. The weight has already shifted, and this deal just externalizes it to those that didn’t realize yet.

To understand it better, we need to understand how the OEM (i.e. the car manufacturers: GM, BMW, Ford and others) ecosystem works, and has worked for many years. In short, an OEM car is the result of an intricate, multi-level tree of suppliers and sub-suppliers, broken down into layers: ‘Tier 1’, ‘Tier 2’ etc. Components go up the tree to the OEM who has full control over the composition of this tree. Money flows down the tree to the suppliers, and everyone are happy.

Enter Data. Data from onboard sensors, cameras and components started playing a role in the automotive space more than a decade ago, but in recent years its importance has grown exponentially. The collection and ownership of driving data has become a strategic advantage, as Tesla has vividly shown.

However, data remained firmly in the hands of the OEMs. They pay for the car components, but keep the data those components produce by-and-large to themselves. They own the full stack.

This is fine if what you’re producing is a driveshaft or an onboard GPS, but it is a huge problem if you’re in the business of developing AI components that aims to warn drivers about road dangers, and have aspirations to own the intelligence stack required for Autonomous Vehicles.

While I have zero knowledge regarding the inner workings of Mobileye, I do know how the industry has changed in recent years, with the introduction and popularization of a set of technologies known as ‘Deep Learning’. Excelling in Deep Learning, a technique which conquered the machine vision profession since 2012, is much less about years of unique IP research & development, and much more about collecting very large unique datasets. A good illustration of this was a recent competition we ran at Nexar challenging researchers to create a traffic lights detector. Within weeks some of the competitors have managed to build state-of-the-art solutions because they had access to a massive real-world driving training dataset from Nexar.

Mobileye has announced partnerships with some major OEMs that have at least some data component in them, and have presented their REM concept which aims to provide high resolution mapping as a service for autonomous vehicles, but the challenge for true autonomous driving is much larger. It requires not only accurate mapping, but rather, the ability to understand and predict what is going to happen on the road around you. You need to train your system to detect vehicles and pedestrians and potholes and anything else.

You need to be able to train your system to the different traits of different cities and different countries. For example, if you want to take your vehicles to India, you need to be able to cope with cows on the road. The task becomes much harder when you move from detecting and classifying objects to detecting and responding to danger. There is a very long tail of potentially dangerous situations happening on the road everyday, and learning all of these situations is a very daunting task.

This complex task requires endless amounts of data, which even the OEMs have a hard time collecting. Getting access to these data sets as a single component manufacture is close to impossible.

The benefit of the Intel-Mobileye deal is that it allows the two companies together to build a more holistic set of solutions that own more of the stack and have a better chance of getting access to the pools of data required for a pole position in this next phase of the industry. It may even allow them to build chips that learn ‘on-the-job’ leveraging the local access to data each car has.

These kind of consolidation deals will be inevitable, as more and more companies realize that playing the component game is no longer an option, and the future margins will come from owning the intelligence stack, which first requires owning the data stack. In fact, I would suggest we should interpret all major upcoming deals in the transportation market through the lens of data and the accelerating race for data ownership.

Thanks to Bruno Fernandez-Ruiz and Michael A. Eisenberg: Six Kids And A Full Time Job.