Intel’s move towards self-driving cars
As PC sales have declined, Intel has struggled. Will self-driving cars bring Intel back to the top?
How Intel is getting involved in self-driving cars
Intel is the world’s second largest semiconductor chip maker, supplying processors to major computer manufacturers. As personal computer sales have started to decline due to increased preferences for mobile devices, Intel has struggled. It’s laid off 12,000 workers since 2016 [1].
The rise of machine learning creates new opportunities for Intel. Driverless cars require substantial processing power, creating opportunities for many tech companies besides auto manufacturers to contribute technology under the hood. According to Bain & Co., this will be a $25 billion market by 2025 [2].
Intel’s involvement to date
Intel has already moved aggressively towards this new market. In March 2017, Intel bought Mobileye, a digital vision technology that helps autonomous vehicles with navigation, for $15.3 billion [2]. This helps broaden Intel’s offering beyond processing chips and positions it as a compelling partner for auto manufacturers reluctant to partner with Google or other tech giants that are potential direct competitors in autonomous vehicles. Self-driving cars require a combination of machine learning algorithms to be successful, from image identification to understand their environments, to decision making and execution [3]. The Mobileye purchase helps Intel cover that first step of machine vision to understand the environment around the car.
In addition to this immediate short-term move, Intel is thinking long-term about building complementary offerings across its business units. Intel’s data center group sells processors and other technologies directly to data centers. This business unit is already Intel’s second-largest, and it’s main driver of growth [4]. It’s also the portion of the company best positioned to grow with the increased prevalence of self driving cars, which require the substantial computing power that data centers provide.
Additional opportunity for Intel: training algorithms with real and simulated driving miles
There are other important areas of this emerging industry that Intel should consider as it positions itself for autonomous vehicles. One key area of consideration relates to the process of helping autonomous vehicle manufacturers train their algorithms. Currently Tesla and Waymo are leading the industry in developing technology for self-driving cars. However, they are doing so using very different methods. Tesla is capturing actual driving data from Autopilot software in its existing vehicles already on the road that customers are driving, capturing over 700 million miles of data. On the other hand, Waymo has created simulated driving environments to generate over 2 billion miles of driving experience to seed the training dataset for its algorithms. Despite this advantage, Waymo engineers acknowledge that at some point, additional driving miles provide diminishing returns to honing the algorithms. What really matters is having unique situations available to train the algorithms [5].
This creates an interesting opportunity for Intel to partner with several manufacturers to help them collect driving data, similar to Tesla’s approach. In January, Intel announced partnerships with BMW, Nissan, and Volkswagen [6]. This might be an angle for Intel to leapfrog Waymo and Tesla to build the largest, and potentially most diverse, training dataset for self-driving algorithms to use.
There are tradeoffs between real miles and simulated miles, which means autonomous vehicles will likely require both sources of training data. Real miles might capture things a simulator might not be programmed consider. Simulations allow algorithms to practice very rare or dangerous traffic incidents. A majority of driving miles that Tesla has captured are mundane, with little marginal benefit to training the algorithms. Simulations can hone in on challenging edge cases that algorithms need to practice. It’s likely that the way autonomous vehicles handle these edge cases will determine public sentiment towards the technology.
As a consequence, Intel should have a strategy to get involved in each form of data collection and processing, including both real and simulated miles. Intel should look to partner with new driving simulator companies to help provide the processing power required to distill learnings from billions of simulated drives. Nvidia is a startup that recently released a ready-made simulator for other company’s self-driving projects. Tesla is using it as well to complement its real driving miles. Once the vehicles drive the training miles, other technologies are required to annotate and segment the driving situations the cars encountered, to identify where to focus training next [7]. If Intel wants to avoid being relegated to a supplier of silicone, it must consider how to participate in each layer of tech required for self-driving cars.
Open questions
Intel’s move towards autonomous vehicles is very exciting and demonstrates that the autonomous vehicle industry creates opportunities for companies to plug in at many different levels, beyond auto manufacturers themselves. Going forward, Intel must consider a few key questions:
- Which capabilities should Intel develop in-house, and which should it look to acquire by buying other startups?
- What can Intel do in this space to create a differentiated and defensible offering?
Word Count: 794
Endnotes
[1] Vanian, Jonathan. “Samsung Dethrones Intel As World’s Biggest Chip Maker.” Fortune, July 28, 2017. http://fortune.com/2017/07/27/samsung-intel-chip-semiconductor/, accessed November 2018.
[2] Scott, Mark. “Intel Buys Mobileye in $15.3 Billion Bid to Lead Self-Driving Car Market.” New York Times, March 13, 2017. https://www.nytimes.com/2017/03/13/business/dealbook/intel-mobileye-autonomous-cars-israel.html, accessed November 2018.
[3] “Unbundling The Autonomous Vehicle.” CB Insights, October 31, 2018. Accessed November 2018.
[4] Eassa, Ashraf. “How Intel Plans to Win Self-Driving Cars.” The Motley Fool. July 9, 2018. https://www.fool.com/investing/2018/07/09/how-intel-plans-to-win-self-driving-cars.aspx, accessed November 2018.
[5] Hawkins, Andrew J. “Inside Waymo’s strategy to grow the best brains for self-driving cars.” The Verge, May 9, 2018. https://www.theverge.com/2018/5/9/17307156/google-waymo-driverless-cars-deep-learning-neural-net-interview, accessed November 2018.
[6] “Intel Partners with BMW, Nissan, SAIC Motor, Volkswagen, Paramount Pictures, Ferrari North America to Showcase Power of Data at CES.” Intel press release, January 8, 2018. https://newsroom.intel.com/news-releases/2018-ces-intel-showcases-power-data/, accessed November 2018.
[7] Anil, Aakar. “What kind of machine learning algorithms do driverless cars use?” Quora, June 19, 2017. https://www.quora.com/What-kind-of-machine-learning-algorithms-do-the-driverless-cars-use, accessed November 2018.
Thank you for sharing this – I find the case study of a previously dominant tech industry giant is fighting to stay relevant quite interesting. To your question on defensibility, one of the areas of defensibility is keeping data proprietary – after all access to good training data is a key competitive advantage for machine learning, especially for computer vision where the data set is more complex and much of the time unstructured. As you have mentioned, aside from the Mobileye acquisition, they are building this defensible ‘moat’ through (hopefully exclusive) partnerships with manufacturers which is a step in the right direction. It will be interesting to learn more how these partnerships will develop in terms of data sharing. Will these suppliers strive to keep the data for themselves as a bargaining chip with other machine vision providers to get better terms? Or will there even be more fundamental issues with data right and privacy regarding the customer? This is especially important when recent developments such as the unauthorized access to Facebook’s user data stirred up such a controversy in recent years and automated vehicles may hold sensitive data such as workplace location and homes. In this case, these partnerships may go back to being primarily a hardware agreement instead of a potentially more lucrative one including data-sharing and creating more sophisticated automation.
I agree that in machine learning, the data is usually the moat, but in the case of autonomous vehicles, there are at least half a dozen organizations that are investing billions of dollars to create hardware and software and build a data set.
Also, at least some of the partnerships with Intel are not exclusive. FCA is working with many partners, one of which is Intel. https://d3.harvard.edu/platform-rctom/submission/the-road-to-autonomous-driving-at-fiat-chrysler-automobiles
Really great explanation of real and simulated miles and how they can be useful.
I posed a very similar question in my essay to the one you posed about how Intel can create a differentiated and defensible offering. It seems to me that is is fairly likely that in 10 years, Level 4 autonomous systems will be relatively commoditized. There are many groups working in this space and unless many of them fall short, there will be many options to choose from. It also seems that there are not many ways to create a differentiated product, especially if all providers get to a relatively similar safety record. The biggest differentiating factor may be in the cost of the sensors and other hardware in the system.
It’s a minor point, but Tesla is generally not considered a leader in the space despite the publicity they garner. This study by Navigant is several months old, but provides an interesting breakdown of where they think each company is: https://www.cnet.com/roadshow/news/tesla-apple-trail-self-driving-pack-study/.