Where Can I Buy a Robotaxi?

---

We’re closing in fast on June…

Tesla’s robotaxis are slated to hit the streets of Austin, Texas, next month in a launch that’s going to completely transform the ride-hailing industry, and for that matter, the transportation industry.

I know it’s something that I write about often here, but it’s impossible to overstate the importance of this launch and the opportunity it presents.

I’ve been researching this technology for more than a decade now… since 2011 when I tested an early prototype of what is now Google’s Waymo.

When Tesla was deeply hated as a car company, I recommended it in 2018 on the grounds that Tesla was actually an artificial intelligence company.

I knew back then that it would be the world leader in autonomous technology, capable of performance far beyond any competitor.  And I also knew that technology would be used to launch an autonomous ride-hailing service that wouldn’t require high-definition mapping or geofencing.  It would work across the country and, eventually – soon – across the globe.

And I leased my own Tesla in 2023 with full self-driving so I could test the development of  Tesla’s full self-driving (FSD) software with each and every software release, experiencing every version firsthand.

And it’s all been leading up to this moment.

One hundred percent autonomous ride-hailing services. Safe, cheap, clean private transportation.

Not like Waymo, which does depend entirely on high-definition maps and geofencing… but a learning, vision-based, extensively built-out full self-driving system with more than 4 billion miles of real-world full self-driving data.

Truly autonomous technology.

The closer we draw to the hinted launch date, the more we hear from readers looking for ways to get involved…

Hello Jeff & Team:

I have been a subscriber for about 6 months now and really enjoy your in-depth analysis of so many different areas of technology, like AI and autonomous vehicles. I took my first test drive of a Tesla in FSD mode last week and was blown away by its abilities, even in rush hour traffic. I am very interested in the Tesla Robotaxi and may want to purchase one, but no one at the dealership has much information on it. They only refer me to the website, which also doesn’t have much information. I know the Robotaxi is due to go live in Austin, TX, in the first week of June, and that’s obviously a big deal. Would anyone at Brownstone know who I might call at Tesla to get more information? Are there other sources of information besides the website that you might refer me to? Any help would be greatly appreciated!

Thank you for keeping me on the Bleeding Edge!

Best regards.

– Herb S.

Hi Herb,

Awesome.  So great to hear you experienced what it’s like to be driven by Tesla’s full self-driving software.  And I agree with you, I think I enjoy it the most during rush hour.  Not having to deal with all the stop-and-go really eliminates any stress from having to drive in those conditions.

And I have good news for you.  You can buy a robotaxi from Tesla.  All you have to do is buy your choice of Tesla and make sure you get the full self-driving software with the purchase.  And voilà, you have your robotaxi.  I’m not kidding…

The initial robotaxi deployments will most likely be Model Ys provided by Tesla.  And then after that, opened up to “normal” Teslas, mostly Model 3 and Model Ys opted into Tesla’s robotaxi network.  In the past, Tesla had inferred it would soon launch a robotaxi app (as shown below), very similar to Uber or Lyft apps.

But some software analysis has indicated that Tesla will be incorporating its ride-hailing app functionality into its normal Tesla app used by Tesla owners.

I’m sure the driving logic is so it’s a seamless experience for the Tesla owner (or lessee) to use their Tesla app to opt their Tesla into the robotaxi network to go make some money.

Now, Herb, there is some nuance here, and I am speculating, but I think I’ll be proven right.

I expect Tesla will launch the robotaxi service in Austin with, let’s say, 20 – 50 of its own cars.  In other words, it won’t allow owners and lessees to opt their cars into the robotaxi network. Not just yet.

This is what I would do if I were in charge of the project. And I would expect that by Q4 or Q1 next year, Tesla will enable the opt-in functionality for people to immediately use their Teslas to earn income.

Herb, this probably doesn’t completely answer your question because I think you might be thinking of the Cybercab with the way you worded your question.

It will be possible to purchase Cybercabs, but it is not possible right now.

Full production is planned for the second half of 2026, and I expect Tesla will start accepting pre-orders early next year after its robotaxi deployments have some miles under the belt and Tesla’s FSD software has been upgraded a few more times.

As soon as I hear that Cybercabs are available for order, I will make sure to give everyone a heads up in The Bleeding Edge.  It will be a phenomenal income-generating machine for those who get in early.

But if you’re keen to invest in Tesla’s robotaxis, even without buying a Tesla, I hosted a briefing back in April where I shared everything we know about the upcoming launch as well as a handful of my favorite companies that I believe are going to soar once the robotaxis start rolling out in cities everywhere.

As I said then, Tesla is, of course, a good way to get exposure to the trend… but the truly exponential opportunity lies in the smaller companies supporting the rollout.

The chips, sensors, data center infrastructure, etc. – the ecosystem that makes the entire autonomous system possible.

I’ll be re-airing that presentation next week, just ahead of the June launch. If you’re interested in learning more, you can go here to sign up for alerts, so you don’t miss it when it’s available again.

How “Humanlike” Is Optimus?

Hi Jeff,

I’m interested in learning more about how Optimus sees, hears, and speaks. For example, do his eyes move like our eyes move? And when he speaks, will his facial expressions seem real?

Thank you.

– Stella R.

Hi Stella,

The quick answer is no, as Optimus is designed today, perhaps to its disappointment, it does not have the ability for facial expressions.

And its “eyes” are, of course, special cameras designed to have a wide field of view, so we won’t see the cameras moving as if they were eyes.

But that doesn’t mean that will always be the case.  Your question actually evokes a major product debate in the industry.

Should humanoid robots be designed to look like “robots,” or should they be designed to look like us with artificial skin and the ability to have facial expressions like humans do?

I wrote about this more extensively in a couple of recent Bleeding Edge issues. In Stab It, It Bleeds, I spotlighted Clone Robotics – a company developing humanoid robots designed to be extremely humanlike. The “world’s first bipedal, musculoskeletal android.”

Its Alpha robot is a little stiff in its movements, sure, but it’s noticeably less mechanical than what we’ve grown used to in early prototypes of humanoid robots. Alpha is designed with more than 200 degrees of freedom and complex muscular, nervous, vascular, and skeletal systems.

Out of curiosity, in that issue, I even polled Bleeding Edge readers on their preferred humanoid robot aesthetic – humanlike or mechanical – which I discussed in that Friday’s AMA – A Robot Kill Switch?

And as you can see in the latest results, opinions are still pretty mixed.

In short, there’s no easy answer.  And the reality is that some consumers will prefer their robots to look as close to what a human looks like as possible, as it will give them comfort, and others will want the clear distinction that it’s a “robot.”

The good news is that the more this technology is developed and built out, the more options there will be. So, no matter your preference (and you can still let us know right here), there will be a bot that fits your needs, preferences, and lifestyle.

And at the rate it’s advancing, I expect it won’t be long.

What About the 23andMe Data Breach?

Hello Jeff,

I just read the bleeding edge newsletter about 23andMe with interest, as my wife and I also spit in a tube a few years back to learn more about our ancestry. I find it curious that you didn’t mention that 23andMe was hacked, which they initially denied, but later had to admit. A lot of the DNA data, including ours, was stolen, and it was one of the factors that contributed to its demise.

Maybe you can explain to us what hackers could do with this stolen information, and if we should be concerned about it. Appreciate your feedback.

Regards.

– Richard M.

Hi Richard,

You’re correct, my data was probably compromised along with yours and your wife’s in that data breach, announced in October 2023.  About 6.9 million accounts on 23andMe were impacted by that hack.

I didn’t leave that information out of Monday’s issue for any particular reason.  I could have easily written five pages on just that topic.  I always try to keep The Bleeding Edge to a length that is easy to digest and fits into everyone’s day.

While I completely agree with you that the hack was a real black eye for 23andMe, the reality is that the company entered bankruptcy because the business model was flawed.

If 23andMe had simply sold its test kits and sequencing at a price that would have generated 60–70% gross margins and not attempted to become a drug discovery company at the same time, it could have been a profitable company.

23andMe’s cash burn was primarily caused by its research and development for drug discovery.  That’s a very different business from collecting spit, sequencing, and producing a report.

As for the hack, it’s important to note that the stolen data did not include the actual DNA records.  Some compromised user accounts resulted in processed reports being stolen, as well as personal information like e-mail addresses, birth dates, ancestry reports, health predisposition reports, and family tree information.

This kind of information would be valuable to cybercriminals for social engineering attacks.  They would have a wide range of very personal data to trick people into believing that the hacker is someone else (i.e., a doctor, insurance company, distant relative, etc.).

To be safe, it’s always smart to never answer a phone call from a number you don’t know, and the same is true for listening to a voicemail (you could be unknowingly downloading malware onto your phone).

The same is true about e-mails.  And you can always click on the “From” e-mail name to see the underlying e-mail address to see if it is suspicious.  Never click on a link in an e-mail from someone you don’t know.

I recommend staying on your toes and being suspicious of anything that just doesn’t feel right.

The Risks of Unverifiable Research Data?

Dear Jeff and Team,

Appreciate you and all that you do! Thanks especially for the recent article on the 23&Me auction sale/acquisition.

I am writing with a question for Friday’s Mailbag. I wondered what thoughts/input you have regarding the recent MIT retraction of the much-lauded paper on AI-accelerated Scientific Research (test-time compute)? Kind Regards.

– Paige D.

Hello Paige,

What an interesting development. I’d bet that we’d learn a lot if we had a conversation with the author of the paper.  Something doesn’t feel right to me concerning this matter.

For anyone interested, Paige is referring to “Artificial Intelligence, Scientific Discovery, and Product Innovation” by Aidan Toner-Rodgers, December 25, 2024.  This paper absolutely took off and got a lot of coverage by major outlets like The Wall Street Journal and Nature.

The paper was designed to study the impact of using artificial intelligence on materials innovation.  What caught everyone’s eyes was that the data set was large – 1,018 scientists at a major corporation (unnamed) – and the analysis was very detailed and thorough.  And the findings were believable in that they were intuitive:

AI-assisted researchers discover 44% more materials, resulting in a 39% increase in patent filings and a 17% rise in downstream product innovation. These compounds possess more novel chemical structures and lead to more radical inventions.

[…]

AI automates 57% of “idea-generation” tasks, reallocating researchers to the new task of evaluating model-produced candidate materials.

[…]

82% of scientists report reduced satisfaction with their work due to decreased creativity and skill underutilization.

It makes sense, right?  Of course, an AI-empowered human will be more productive in scientific discovery.  And of course, it will shift the kind of work that a human researcher will focus on.

And it’s probably intuitive that some scientists will feel reduced satisfaction with these newfound tools.  That great feeling of satisfaction when we “figure it out ourselves” isn’t the same when using AI.

But, before we go too far, it does appear that the data set used by Toner-Rodgers was deeply flawed and manipulated, presumably to get such clean results as were in the paper.

The MIT economists named by Toner-Rodgers in the paper – that’s Daron Acemoglu (a Nobel Prize-winning MIT economist) and David Autor (an MIT labor economist) – both lauded the paper when it came out.  Acemoglu said, “It’s fantastic,” and Autor said, “I was floored.”

And yet, a few days ago, both “experts” published a request for withdrawal on the basis that they had no confidence in the veracity of the data or the research.  What a black eye for MIT, Acemoglu, and Autor.

What I did appreciate was that someone from the industry contacted Acemoglu and Autor, explaining why the research didn’t make sense.  After all, someone who works in materials science (industry) has a very different perspective than two MIT academics who are economists, nonetheless, not materials scientists.

That’s why they were forced to take the paper down.  And Toner-Rodgers is no longer at MIT, which clearly suggests that there was fraudulent activity regarding the research.

Also worth noting is that Corning (GLW) issued a complaint against Toner-Rodgers for domain name infringement; you can see that here.  Toner-Rodgers had registered the domain www.corningresearch.com, which suggests that the source of his data/research had been with Corning.

Putting everything aside, the whole incident could have been avoided simply by having a few materials scientists review the paper in advance of being published.  They would have caught the fraudulent approach, and that would have been the end of it.

The veracity of datasets for scientific research is particularly difficult to solve for.  And unfortunately, manipulated data designed to optimize results is far too common.

I’m reminded of “Climategate” when hacked e-mails in 2009 revealed that researchers from the University of East Anglia in the U.K. had falsified key data to exaggerate global warming.

It was a horrible black eye for the entire movement because most global warming research was based on that research, which was based on falsified data.  It was this event that led to the change of terminology from “global warming” to “climate change.”

Data provenance, or data veracity, can be an easy or difficult problem to solve for, depending on the application.  After all, someone, or some team, has to vouch for and prove the validity of the data.

With regards to scientific research, this is typically done by an academic institution or a corporation.  Both have conflicts of interest, and both have been known to falsify data.  The best methodology for confirming veracity is peer review.

Registration of datasets and/or research confirmed through peer review is a useful mechanism to reduce these kinds of instances.

For something like property rights, the problem is much simpler.  Provenance can be proven legally, and then, ideally, registered on a public blockchain so that repetitive searches are no longer required.

Better yet, if it is intellectual property, use of said property can be transactionalized using smart contracts and digital assets on a blockchain.

But with all this said, I’d be remiss if I didn’t make a key point.

Trillions of dollars are being invested in the private sector in bleeding-edge technology right now.  Academia is moving at a linear pace.  Industry is moving at an exponential pace.

What most will see won’t be academic papers. They’ll see the new product or service that is changing the world.  And most data sets that are used to develop new products by tech companies that give that company a proprietary advantage won’t need peer review.  They’ll remain in-house, under high security.

Acemoglu has been critical of AI, his paper “Harms of AI” written in August 2021 took a damning position against the technology.  I won’t bore anyone with his drivel, but if you’re interested, you can see it here.

Autor has also been critical of AI’s future impact on the labor markets.

The key point is that all the politics, corruption (to get grants), decels, and falsifications in academia mean less and less with each month that passes.

Academia has lost its high perch.  The real turning point was the pandemic, when the most prestigious universities dropped evidence-based scientific research in exchange for pushing a political narrative so that they could receive massive U.S. government grants.  MIT, Harvard, Yale, and so many others are guilty of that.

Industry doesn’t care.  It just accelerates.

And the pace of technological innovation will never be slower than it is today.  Each month brings more acceleration.

E/ACC,

Jeff