Tesla Vehicle Deliveries and Projections

Last Updated: July 25, 2018 Originally Posted: June 12, 2018

In analyzing the Tesla subset of the MIT-AVT dataset, I came across the need to estimate how many Tesla vehicles have been delivered, and how many of them have Autopilot and which version of Autopilot hardware. I organized the data from various sources (CSV shared below) and made a couple of plots. Also, I did a conservative projection out to January 1, 2021 based on continued average weekly delivery rate that was achieved in 2018-Q2. If you find any inaccuracies in the data please let me know. is this tracked? This blog post is continuously updated, see notes below for changes with each update.

If you have questions, suggestions, requests, connect with me on social media: TwitterLinkedInInstagramFacebook, or YouTube.

The basic timeline of relevant model and Autopilot hardware milestones:

  • Jul-2003: Tesla founded.
  • Feb-2008: First Tesla Roadster delivered.
  • Jun-2012: Tesla Model S released.
  • Sep-2014: Autopilot hardware 1.0 installed (Autopilot not enabled).
  • Sep-2015: Tesla Model X released.
  • Oct-2015: Autopilot enabled.
  • Oct-2016: Autopilot Hardware 2.0 released.
  • Jul-2017: Tesla Model 3 released.

The primary source for the data that follows are the quarterly investor letters. We scraped these to generate a quarterly estimates of Tesla vehicles delivered (by model) with minimal interpolation except before 2012 when information on Roadster deliveries was sparse. You can download the CSV with this aggregated data here: tesla_vehicles.csv. The following is the plot showing the total number of delivered vehicles, segmented by Autopilot hardware.

Two notable observations:

  1. There are over 269K Autopilot-enabled Tesla’s on public roads.
  2. There are now more Autopilot hardware version 2 cars delivered than version 1.

Vehicle Deliveries: Past, Present, and Future

I made an animated plot that shows growth of Tesla deliveries over time and projects them forward to January 1, 2021. The plot has a simple projection model based on the number of vehicle deliveries in the most recent quarter:

Projection Model: Maintaining an unchanged average weekly delivery rate of 4,387 Model S, X, and 3 vehicles. This was the average delivery rate achieved in 2018-Q2. This estimate is potentially very conservative since it assumes that neither the production rate nor the delivery rate increases at all from the previous quarter’s average.

Side note: Currently production rate is higher than delivery rate. This is not due to insufficient demand (as far as has been reported) but rather to the challenges of delivery pipelines supporting the rapid ramping up of production levels. As much as possible, I try to focus on providing numbers on deliveries and not just production. The former is what ultimately defines the point at which we can start counting miles.

The following is a still image version of projections as of January 1, 2021, showing total vehicle deliveries of 934,181 if Model 3, S, and X delivery rates remain at 4,387 a week. Once again, this estimate is potentially very conservative since it assumes that neither the production rate nor the delivery rate increases at all from the previous quarter’s average.

Obviously, prediction of the future (no matter the model) is impossibly difficult, and should always be taken with a grain of salt. Also, I’m an AI researcher and not an economist nor a financial analyst. My interest in Tesla is purely in that it’s one of several implementations of human-robot interaction in the real world. It’s a fascinating and important interaction to study.


Based on helpful discussion, I’ve made several updates to the data, the plots, and this blog post since first publishing. I summarize the updates here.

  • Update 3 (July 25, 2018): I updated the plots based on the 2018-Q2 delivery report from Tesla. Also, I removed any projection model that is based on “expected” production/delivery rates. While those are quite likely, based on most reporting I’ve read, my research is focused on understanding human behavior in the context of Autopilot and not financial investment, so a conservative estimate of future deliveries is sufficient.
  • Update 2 (June 12, 2018): I included Model S/X in the projection of the total number of vehicles delivered, assuming that average Model S/X production remains the same as it was in the first quarter of 2018.
  • Update 1 (Jun 11, 2018): I changed the projection models considered to ones based on Model 3 current and expected Model 3 production levels.


Based on questions I received, here are some comments:

  • Q: How did you make the animated plot?
    I wrote a wrapper around Matplotlib that allows for animation of time series data.
  • Q: Your projection models don’t include any longer-term ramp-up of production. How come?
    A: I chose to only project forward based on current or near-term (within a month) publicly announced rate expectation (projection A and B, respectively). Anything higher than that, while possible, is more speculative than I’m comfortable with. So you can think of the two projection models as conservative estimates of what is achievable.
  • Q: Are you assuming projections in production reflects the projection in demand?
    A: Yes, good point. For the date range of the projections, the assumption is that demand will match supply. There is not guarantee that this will be true.
  • Q: Do you work for Tesla? Do you own Tesla stock?
    A: No. I don’t own Tesla stock. I don’t work for Tesla. I’m a research scientist at MIT who (among other things) builds and studies AI systems that interact with humans. I study human behavior in the context of Tesla Autopilot to understand how we can make semi-autonomous system safe and enjoyable to use. I do my best to be objective, and let the data do the talking. To state the obvious, there is no amount of money which can purchase my opinion, my analysis, or my rigorous exercise of the scientific method. I do what I do not for money (again, obviously) but because I have an insatiable curiosity about human nature and artificial intelligence, and a desire to build systems that help save lives and improve the quality of life for as many people as possible.
  • Q: I have a Tesla, can I join your study?
    A: Sure. First, I’d appreciate it if you filled out this Tesla Autopilot survey. Second, to learn more about the data in the study see this page. Third, to let us know you’re interested go to this page.


The important takeaway here is not the future but the present as captured by the data that represents what has already been achieved. Tesla has put over 310K Autopilot-enabled Tesla vehicles in the hands of consumers, and that to me provides a powerful opportunity to study and understand how AI can help save lives through successful, long-term human-robot interaction, communication, and collaboration.

That’s exactly what we’re doing with MIT-AVT study. The fundamentals of human-robot interaction with Autopilot we’re observing are fascinating. As a closing request for aiding our research: if you’re an owner of a Tesla or know someone who is, please consider taking/sharing our detailed survey on Autopilot.