Introduction: data vs. algorithms

Peter Norvig, a machine learning expert and director of research at Google (GOOG, GOOGL), famously said of Google’s web services: “We don't have better algorithms than anyone else. We just have more data.”

In 2015, Google open-sourced its machine learning software library TensorFlow. Technology and strategy analyst Ben Thompson argues that Google can afford to give TensorFlow away (and accrue the benefits of doing so) because Google’s competitive advantage lies in its data and its massive data processing infrastructure. In this case, Google has the same algorithms as everyone else — the ones it has given away for free — and retains its lead through data.

Facebook (FB) and Amazon (AMZN) have both open-sourced some of their machine learning algorithms as well. This further reinforces the point that competitiveness in machine learning is more about data than about secret, proprietary algorithms.

In 2015, Tesla (TSLA) CEO Elon Musk and Y Combinator President Sam Altman co-founded a non-profit called OpenAI. With $1 billion in funding from various donors, including Musk and Altman, OpenAI brings together top researchers to push the state of the art in machine learning and to make their algorithms freely and openly available to the world. While OpenAI’s stated mission is philanthropic (to democratize AI), as Ben Thompson observes, it also has a benefit for Tesla and Y Combinator, as well as the companies of other donors.

OpenAI ensures that state-of-the-art machine learning algorithms continue to be open source, and spreads the cost of research and development across multiple companies. As a result, Tesla can remain competitive with world-class AI companies like Google as far as algorithms go, without relying on their generosity in open-sourcing their own algorithms. Tesla also avoids the expense of attempting to develop algorithms by itself, while OpenAI’s philanthropic mission can also help attract top talent.

As a bonus, it turns out that Musk’s role as co-founder, co-chair, and major donor to OpenAI gives Tesla access to some of that top talent as well. Tesla hired its new Director of AI, Andrej Karpathy, from OpenAI.

So, bringing all of these pieces together, it seems that a company’s competitive advantage in machine learning will be determined primarily by its data, not by its algorithms.

Tesla’s driving data

This is as true in self-driving as in any other domain of machine learning. Given that different companies appear to be about on par in terms of algorithms, what will differentiate the performance of their self-driving software will be data.

Whereas the self-driving test fleet of Waymo (formerly the Google self-driving car project) comprises around 700 or 800 cars, Tesla uses its Hardware 2 production cars, driven by customers, to collect driving data and to test and validate its software. Tesla currently has about 70,000 Hardware 2 cars on the road, and adds about 2,000 per week. Tesla is currently ramping up production of its mass market car, the Model 3. By the end of the year, it plans to produce an additional 5,000 cars per week (or 7,000 total) and an additional 10,000 by the end of 2018.

A Tesla Model X. Photo credit: Tokumeigakarinoaoshima.

This means that Tesla currently has approximately 100x as many vehicles on the road collecting data as the company with the second largest fleet. As Model 3 production ramps up, it plans to grow its fleet another 10x by mid-2019. That’s in addition to whatever number of internal test vehicles Tesla may have. The amount of data Tesla has access to is unparalleled.

This level of data collection makes Tesla’s aggressive timeline for self-driving credible. Recently, Musk said that Tesla may develop the technical capability for full level 5 autonomy as early as 2019.

The economic opportunity and necessity of self-driving

Technology think tank RethinkX predicts that by 2030, 95% or more of U.S. passenger miles will occur in self-driving electric cars.

 Source: Rethinking Transportation.

The tech-focused investment firm ARK Invests predicts that ride-hailing services making use of self-driving cars will generate $10 trillion in annual revenue globally in the early 2030s.

Source: ARK Invest.

There are therefore two standpoints from which to view self-driving. Developing self-driving cars is both a necessity in order for an automaker to stay in business over the long term. It is also potentially an immense revenue opportunity for an automaker that launches its own ride-hailing service using self-driving cars, as Tesla plans to do with the Tesla Network.

Investors should therefore find Tesla’s data lead attractive, and should also avoid investing long-term in automakers that don’t have a clear strategy for self-driving.