Tesla leadership is chaotic, drove away the top talent, did serious brand damage, overpromises and under delivers repeatedly, and inserts itself into implementation details that results in cl*sterfudge such as Cybertruck and camera-only/mostly self-driving with about zero L4 customer miles with Tesla assuming liability and declaring hardware as “L4 autonomy ready” when it wasn’t.
That’s enough, I’d say.
If Tesla can continue to develop the EV market and improve driving safety in a material way, then I’m very happy for them to achieve that.
I hope everyone makes profitable trades, regardless of business performance.
Why would there be so many departures over the last year or so when they just launched robotaxi to 50% of the US?
Weird.
Sendil Palani, Tesla’s Vice President of Finance, announced his departure from the electric vehicle maker today, marking the end of a nearly two-decade tenure that spanned the company’s transition from a struggling startup to a global powerhouse.
Thomas Dmytryk, the director who led the team that built Tesla’s over-the-air update infrastructure and the software backbone of its Robotaxi ride-hailing service, has announced his departure after 11 years at the automaker.
TSLA stock is more a bet that the (undeserved) Musk worship will continue, rather than a bet on actual fundamentals. When was the last time the company actually delivered on a promise? Give me the rest of the Mag7 any day, companies with strong leadership and reasonable valuations.
Um, how about the one where they said they’d have automous vehicles in commercial service? Which they did, and do, in Austin.
There are dozens if not hundreds of predictions (it’s wrong to characterize them as “promises”) that Tesla has achieved since it was founded about 20 years ago. From Roadster onwards. I remember one where a loud-mouthed critic promised to eat his hat if a Tesla vehicle could change lanes on its own. And when that happened (in production vehicles), the hat went un-eaten.
Do not confuse prediction timelines with prediction achievements. While certainly not 100%, Tesla and Musk have achieved many things they told us they were striving for, including things that many thought couldn’t be done, and things that other companies didn’t do until afterwards, or even still haven’t done. That the prediction timelines from Musk have usually been wrong isn’t great, but considering the ground-breaking nature of some of these achievements, that Tesla was still able to achieve them before anyone else is remarkable.
A common way to value a business is the sum of future cash flows discounted back to the present, using the formula: (1 + r)^n
where r is the discount rate and n is the number of years.
For example, if cash flows are expected to arrive in Year 1 but are delayed to Year 3, the cash flows are discounted two more times.
There are other costs that aren’t included in the equation, like continuing R&D and operational costs before the cash flows arrive, loss of market share, etc. This means when the cash flows arrive matter a lot when it comes to valuation.
The Tesla bull case often includes discussion of Optimus and robotaxi as justification for the high P/E. Tesla was supposed to have produced a “legion” of Optimus by the end of last year. Now it looks more like then end of this year. That lowers the valuation a lot. And of course, the delay gives more time for the competition to enter.
Same with robotaxi. You explained in another post that Tesla’s strategy in California is to wait for the Federal government to act, then and apply for ride hailing permits. The Federal government is not known for its speed. A version of the SAFE DRIVE Act now in committee first passed the House in 2017. If the current bill gets out committee, it must pass the full House, and then the Senate, and then NHTSA has to promulgate the final rule. So we can safely assume nothing will be final until next year at the soonest.
And of course, outside of Texas, Tesla does not appear to be applying for AV ride hailing permits in any other states, apparently following their “California” strategy of waiting for the Federal government to act first. This changes the robotaxi rollout timeline by a lot. Tesla’s competitors in this space are not waiting, however.
Tesla is still by far the US leader in EVs and still has good margins for a car company. But the P/E of 350 is misplaced if people are counting on robotaxi or Optimus revenues anytime soon.
I would absolutely short TSLA based on fundamentals, but a cult stock can continue upward for longer than I can stay solvent. The meme game works for some people for some time. But as an actual investment, this is a sure loser longer term.
Same reason people who got rich on AOL got poor a decade later? Same reason AIG investors who got rich in the aughts got poor after 2008? Same reason people who got rich on Cisco did a flatline for 20 years after the 90’s? Same reason people invested in RCA in the 1920’s and then fled in the 1930’s?
Because the future doesn’t always reflect the past?
Because the CEO is distracted?
Because the car sales which account for 80% of revenues have declined for several years in a row?
I dunno. Pick one from Column A and one from Column B.
We know from Tesla’s current lack of L4 vehicle autonomy that they haven’t demonstrated meaningful capability and reliability.
Here are specific reasons to doubt their near-term (2026, 2027, 2028) technical ability to achieve meaningfully in-house L4 vehicle autonomy.
Tesla lost substantial senior talent over the last several years.
From Waymo’s autonomous vehicle (AV) rollout we have evidence that multi-sensor works, but there is no such evidence for camera-only/mainly sensor configurations (Tesla’s approach).
Other AV competitors, beyond Waymo, are mostly multi-sensor.
There is a high probability that Waymo evaluated camera-only and determined multi-sensor is the better first pass design. They have subsequently reduced their sensor count in newer generation hardware.
We can assume Tesla is at least one order of magnitude away from human or better safety levels (accidents per 1 million miles). Data on scaling laws from Waymo suggest diminishing returns to safety improvement with each 10x increase in compute power (and model parameter size), and Tesla would take at least one year to iterate a large software (and maybe hardware) version change (eg, software v13 to v14, and next to v15).
If we just extrapolate Tesla’s recent historical improvement in safety, they have at least a year or two to go to get to human safety or better.
No California AV testing by Tesla is happening, suggesting their safety level is not good enough for them to feel ok about disclosing it to the public.
Tesla needs at least 20 million L4 autonomous miles to start to statistically demonstrate human level safety. This will take time and they haven’t meaningfully started. Waymo has passed 200 million autonomous miles.
Tesla has admitted that a prior vehicle hardware version, HW3, cannot support unsupervised driving software. There are strong indications that the capability of the latest hardware, HW4, has plateaued and also will not achieve autonomy at better than human safety. This would mean that the next generation, HW5, or an even more advanced version, will be needed to deliver autonomy. HW5 is in the process of being manufactured and is likely at least a year away from being used in any meaningful quantity.
More advanced sensors (eg, automated cleaning) have been seen in Tesla’s robotaxi fleet, but not in customer vehicles, suggesting additional hardware changes beyond more advanced chips are needed for autonomous driving to handle various environmental conditions (water, dust/dirt, ice, etc). It will take additional time to develop, test, and manufacture this additional hardware.
Tesla could partner/acquire the needed technology, which is a different path that could help overcome some of the above doubts.
I’m replying mainly because these boards have been dead way too long.
This post by “mostlylong” was so exceptional that I thought someone might try to refute it it, but nobody has. I wonder why? I guess maybe everyone understands how accurate it is?
IMO, I actually think mostlylong’s points may be somewhat conservative. Tesla’s robotaxi hurdles now to overcome may be even more difficult on a couple points.
First, we STILL have no evidence that gves a high confidence level that camera only will ever work.
Secondly, rather than say, “Tesla is at least one order of magnitude away from human or better safety levels”, in my opinion it is clearly more than just one order of magnitude. Is it as many as 3 or maybe even 4?
Rather than, 'Tesla needs at least 20 million L4 autonomous miles to start to statistically demonstrate human level safety." I have never seen anything that really gives confidence based on testing or actual unsupervised robotaxi miles that a robotaxi can then be deemed safe. Obviously Elon claimed the Billions of miles of raw data would do the trick then he raise3d that to 10 billion. As the 10 billion miles is close, the incidents of FSD problems still remain. While issues have continued to arise with paid, unsupervised driving I just don’t think 200 million miles will mean anything for Tesla even if they reach it soon.
Lastly, we really don’t know exactly how bad the hardware problem is. Maybe HW 5 will be better, or even the so-called HW4+ – what we do know is that this limited testing in various locations will continue as is but is it moreso PR or is it really proving/disproving anything? Way too many people think the lack of alternative sensors will continue to be a problem and we are still told that tesla rejects that.
Again though I commend mostlylong for his excellent post
I agree, I believe they are conservative so I used phrase “at least” a few times.
On this one:
Let me explain further.
Let’s distinguish between the number of miles to train a model versus, given a trained model, the number of miles to demonstrate (measure) safety at human level or better by that (already) trained model.
I am discussing the latter.
Here is a thread on that topic:
and the Waymo study on same:
Finally,
One challenging item is that the AI driver could be 3 or 4 nines good: 999 or 9,999 miles without an accident-preventing human intervention.
But this is still terrible: 1 accident per month or year, roughly (details depend on exact definition of accident severity, more severe accidents are rarer).
But to the average person driving FSD, this might seem really good: they only intervene 1 to a few times per year.
Of course, are people really letting FSD drive 100% of their driving miles? And any route they would normally take? Almost any environmental conditions/weather/time?
I have a follow up question on the above. It seems when there are references to “march of 9s” where I see several different ways of usage for the term. You mention “miles without an accident-preventing human intervention”; I mostly see these as %'s. As one example 99.99% “safe” would be 4 9s and same as 1 accident in 9,999 miles as your comment. It seems there are even other variations though — do you know if there is an “official” explanation for these?
I personally think the community sourced FSDTracker is very useful to check. It shows based on supervised driving - by participating owners/drivers - without critical errors expressed as %. Of course these are different as it is supervised and critical errors is a broader category than “crashes”. Because these are supervised, the driver reports incidents where he/she prevented a crash.
Despite that; these are meaningful at least to me. The very blatant statistic in virtually every release is that I have never seen a case for any release - once meaningful mileage had been reported - where the % even made it to 99.9%. With significant miles accumulated already the newest V14.2.2.4 release only shows 98.9%. Again, this is not an apples to apples compare but still very useful. It doesn’t seem to matter much to me if a supervising driver had to intervene to avoid a crash though - perhaps that would have happened n some portion of these.