A conversation with Google AI

Stocks A to Z
1

These are results for some investors don’t seem to understand Tesla’s pivot from passenger cars to humanoid robots and cybercabs

GoogleAI:

Tesla is undergoing one of the most polarizing transitions in corporate history, shifting from a high-volume EV manufacturer to an AI and robotics ecosystem. While long-term bulls view this as a multi-trillion-dollar unlock, traditional auto investors often struggle with the pivot because it replaces predictable manufacturing metrics with highly speculative, long-horizon tech valuations.

To understand why this friction exists, we have to look at how different investor classes value the company.

The Valuation Rift: Auto Metrics vs. AI Horizons

The primary source of confusion stems from a fundamental mismatch in valuation frameworks.

  • The Bear/Value Case: Traditional analysts value Tesla on automotive gross margins, delivery growth, and free cash flow. From this lens, slower EV demand and price cuts signal a maturing car business that cannot justify a tech-premium price-to-earnings (P/E) multiple.
  • The Bull/AI Case: Growth investors view the car business as a mere hardware distribution network meant to fund and feed data into the true value drivers: Full Self-Driving (FSD), Cybercabs, and Optimus humanoid robots.

Breaking Down the Pivot

The strategic shift relies on three deeply interconnected pillars, each carrying a vastly different risk-reward profile:

[EV Fleet Data Engine] ──> [FSD & Cybercab Network] ──> [Optimus Robotics Platform]

1. Cybercabs & Autonomous Transport

  • The Strategy: Transitioning Tesla from a car seller to an autonomous ride-hailing network operator.
  • The Friction: Regulatory hurdles, localized mapping approvals, and legal liability in “unsupervised” FSD mode remain unproven. Investors fear a long, capital-intensive road to true Level 4 autonomy.

2. Optimus Humanoid Robots

  • The Strategy: Deploying thousands of Optimus robots internally to lower Tesla’s own manufacturing costs, before selling a generalized labor workforce to external factories and warehouses.
  • The Friction: The addressable market is theoretically massive, but commercial scalability, dexterity limitations, and unit economics are years away from being validated.

3. The Core EV Anchor

  • The Strategy: Utilizing the existing global fleet of millions of customer vehicles as a real-world data engine to train the neural networks powering both the cars and the robots.
  • The Friction: If core vehicle margins decline too sharply, the capital required to fund these massive AI compute clusters (Dojo and Nvidia H100 networks) risks squeezing Tesla’s balance sheet.

Structural Risks to Keep in Mind

If you are evaluating Tesla’s current trajectory, consider these wealth-degrading mechanics and structural risks:

  • Capital Reinvestment Risk: Diverting billions from affordable next-gen passenger cars to robotics risks losing EV market share to aggressive global competitors before the AI bets mature.
  • Timeline Compression: AI and regulatory timelines are notoriously difficult to predict; a multi-year delay in Cybercab commercialization could trigger severe multiple contraction.
  • Concentration Premium: Holding Tesla right now means investing in a concentrated bet on Elon Musk’s execution of Artificial General Intelligence (AGI), rather than a diversified auto play.

What I have discovered is that while waiting for the future selling covered calls is a nice income generator. In addition to Tesla I’m also betting on AI datacenter providers.

The Captain

1 Like
2

Captain,

I know you love your AI responses, and even though I don’t find them super useful or interesting to read, it’s certainly your prerogative to use and post them, if you like. But I think there are two things you should keep in mind.

First, the AI is going to give a response based on what you actually prompt it - so it will reflect back to you the reality that you put into it. Here, your input starts with the premise that some investors don’t understand Tesla’s pivot. The output will take that as the premise, and not address the actual argument against this pivot by investors who do understand the pivot, but have concerns about it. So if you want to get useful information out of the AI, you absolutely need to pay attention to the specific language you use in your prompts, because the very specific words you use in the prompt will affect your output.

Second, and most importantly, don’t ever forget that AI learns about you from all your past inquiries, and that it has a primary goal of keeping you happy and engaged with it. So the response you get will be tailored to you. It will shape its version of the response to match what it thinks you will be most receptive to. When I enter the same prompt in my own version of Gemini, I got a somewhat different version of the response:

Tesla’s pivot from passenger cars to artificial intelligence, humanoid robots, and cybercabs has created one of the sharpest divides on Wall Street. Skeptical investors are struggling with the transition because 95% of Tesla’s current revenue still comes from selling traditional electric vehicles, a core business that has recently experienced plateauing sales and falling automotive profits. By shutting down classic lines like the Model S and X to fund a $20 billion annual AI and robotics capital expenditure, Tesla is essentially trading stable, near-term manufacturing cash flow for highly speculative, unproven future industries. [1, 2, 3, 4, 5, 6, 7]

The disconnect centers on a fundamental clash between traditional valuation metrics and long-term tech speculation:

  • The Valuation Strain: Traditional investors see a company trading at nearly 200x forward 2026 EPS targets. For an automaker, that valuation is entirely disconnected from reality; however, Elon Musk has openly stated that about 80% of Tesla’s long-term value will come from the Optimus robot. [1, 2, 3]
  • The Cybercab Realities: While bulls focus on a potential $1.4 trillion autonomous mobility market, skeptics look at immediate regulatory roadblocks. The camera-only Cybercab lacks manual controls (no steering wheel or pedals), meaning it faces intense regulatory scrutiny from agencies like the NHTSA, restricting it to tightly geofenced testing areas for the foreseeable future. [1, 2, 3]
  • Immediate Financial Drag: Reinvesting all free cash flow into training robots and building Dojo supercomputers has hit Tesla’s immediate balance sheet. Analysts worry that meaningful revenue from Optimus or driverless ride-hailing networks is still years away, leaving the company exposed to a collapsing near-term EV market. [1, 2, 3, 4, 5]

Ultimately, the market is split between evaluating Tesla as a legacy hardware automaker or a future AI ecosystem. [1, 2]

Note how my version differs from yours - it’s much more skeptical in tone about Tesla, no doubt because many of my past queries to Gemini have been pushing at the edges of Tesla’s valuation and growth story, and less looking for support for a current investment holding. Gemini certainly knows you own Tesla, probably knows that you own a lot of Tesla, that you prefer information that supports (or at least elaborates on) the bull case for Tesla, etc. It’s going to tailor its answer to you with all that in mind (and my version to me, conversely).

Enjoy interacting with our future AI Overlords (Blessed be their glittering server racks), but be careful out there!

5 Likes
3

The AI responses are too long, they are not reliable, they reflect the bias of their prompt from biased people who post them.

At one time I was told TMF was going to simply ban AI generated responses. It hasn’t happened but I think it should. It seems they add nil value except to those that post them.

6 Likes
4

Don’t switch this thread to be about me. What’s your response to THIS AI output?

The Captain

5

I didn’t find it useful or interesting. I don’t think there’s too many investors that don’t understand Tesla’s attempt to pivot from passenger cars to humanoid robots and cybercabs. They know what Tesla’s trying to do. They just think it’s pretty unlikely that those efforts will pay off for Tesla, certainly not any time soon and probably never.

It’s not that they don’t understand it. They just think that cybercabs and humanoid robots are just Solar Roof all over again - products that very well might exist in widespread adoption some day, but are unlikely to ever make Tesla much money, and so don’t warrant assigning much value to.

Your AI output just reflected back what you told it you wanted. You told it that investors didn’t understand the pivot, so that’s what it told you back. Again, not very useful or interesting.

3 Likes
6

How would you pose the question in neutral terms. I’d love to try it.

The Captain

7

What’s the question you want to know the answer to? I genuinely don’t know what you’re trying to get out of AI. I think you already understand why investors disagree on Tesla’s current valuation - some investors think they’ll be successful with their new ventures, others don’t. What do you want to know from AI about that?

1 Like
8

What I posted at the top of the thread. How to ask this question in neutral terms:

Some investors don’t seem to understand Tesla’s pivot from passenger cars to humanoid robots and cybercabs.

The Captain

9

But that’s not a question. It’s a statement. That’s why I’m confused.

What is it that you are trying to find out?

1 Like
10

WHY → Some investors don’t seem to understand Tesla’s pivot from passenger cars to humanoid robots and cybercabs.

What are they missing? Why don’t they get it?

Couch that in neutral terms.

The Captain

11

You can’t couch that in neutral terms, because it’s not a neutral question. You’re assuming that investors don’t understand Tesla’s pivot. That there’s something that they’re missing, that there’s something they don’t “get.” If you ask that question of an AI, it will just mirror your worldview back to you.

In reality, investors do understand the pivot. They just don’t all agree on whether Tesla will be successful in the pivot, or whether the pivot will yield a highly profitable or largely immaterial stream of income, or how long the pivot will take.

IOW, if you present AI with a prompt that tells it that the reason some investors value Tesla lower than others is because they don’t understand the plan - rather than that they believe the plan has a higher chance of failing - then AI will just reflect back what you told it to assume.

To avoid that, you could simply ask the question, “Do investors understand Tesla’s pivot from passenger cars to humanoid robots and cybercabs?” That’s a question that inputs into AI something that you don’t know for certain, rather than telling it the answer up front. When I asked this of Gemini, it responded:

Yes, absolutely. For the vast majority of regular investors, the issue is not a lack of understanding—it is a fundamental disagreement over execution, timing, and profitability.

Regular investors fully grasp the concept: Tesla wants to stop being a low-margin hardware car manufacturer and become a high-margin autonomous network and robotics provider. They get the vision. They just don’t buy that it will work out the way Elon Musk promises.

That’s the result of a more neutral way of framing the initial question. You could also ask, “Why do some investors value Tesla’s cybercab and humanoid robot ventures more higly (or lower) than others?” if you’re genuinely interested in getting some “why” answers - but if you tell the AI that the reason is because they don’t understand it or they don’t get it, that’s what AI will tell you back.

But please see my earlier post on AI learning about you. You use Gemini a lot, and it has a vast amount of data on you. Even if you strip out the presuppositions in the questoin prompt, you will probably get answers that reinforce the presuppositions anyway. Because even though the “Gemini” is the same, one of the steps Gemini takes in responding to a prompt is to look at all your prior chats with it to garner information about how to tailor its response. It knows you. It already knows that you think that investors don’t understand Tesla’s pivot, so even if you ask it about the pivot in a neutral manner, it’s going to know that a response that presupposes lack of investor understanding is more likely to get you to keep using the tool than if it gives you the sort of answer it might give someone without that background.

If you were actually looking for a neutral, unbiased response, you’d have to ask it of Gemini in a completely new account. Or precede your question with tons of very specific prompts directing the AI tool not to look at what it knows about you in preparing its response (and even then, no guarantee it will obey those directions).

2 Likes
12

You say it can’t be done… :sad_but_relieved_face:

Then you give the answer I was looking for. :smiley:

Here is what GoogleAI replies to your query:

Investors clearly recognize Tesla’s strategic shift toward autonomy and AI, but they remain deeply skeptical about whether the company can successfully execute this high-risk pivot. [1, 2]

:bar_chart: Recognizing the EV Margin Squeeze

  • Chinese automakers like BYD dominate mass-market electric vehicles.
  • Traditional vehicle profit margins are shrinking rapidly.
  • Tesla is phasing out older Model S and X production. [1, 2, 3, 4]

:warning: Doubts on Execution and Timelines

  • Cybercab projects face regulatory hurdles and delays.
  • Meaningful revenue from Optimus robots remains years away.
  • Competition from established robotaxis like Waymo is fierce. [1, 2]

:chart_decreasing: Soaring Costs and Valuation Pressures

  • Capital expenditures surged past $20 billion for the year.
  • Free cash flow turned negative due to heavy AI reinvestment.
  • Stock trades at an inflated forward P/E near 200x. [1, 2, 3, 4]

This is what Fools have been preaching. I can claim that AI has been corrupted by Fools and we’ll get nowhere.

In a reply to Goofy I stated the reason the pivot is a good idea, shifting from EV’s Decreasing Returns (the Chinese are coming for you) to taxis’ and robots’ Increasing Returns. Amazon is a great example, from Decreasing Returns selling to Increasing Returns AWS. That’s the whole point that doubters miss. Of course it’s not a certainty but not going for it dooms Tesla to become an also ran.

Notice that I dismissed all the negatives AI came up with because those are the hurdles the pivot seeks to avoid. Chinese EV competition, etc.

o o o o o o o o o o o o o o

Your AI concluded:

Elon hate? Elon envy? Elon politics? Elon free speech? The world’s current most successful industrialist a failure? Since Tesla is going broke we might as well short the stock! :slightly_smiling_face:

Not interested. Read my post about how intelligence works.

The Captain

1 Like
13

Doubters don’t “miss” it. They know that’s what Tesla is trying to do. They just think it’s not going to happen. They don’t think taxis and robots will offer increasing returns, because they don’t think Tesla is going to be able to develop economically viable versions of either of those two products.

If what you want is to genuinely understand why some people are skeptical of Tesla’s strategy in pivoting, you probably should move past the idea that it’s because they don’t understand it, or that they miss why Tesla is doing it. They do. They understand it, and they know why Tesla is doing it. They just disagree with you (and Tesla) about: i) whether or when those products will be technologically feasible; ii) how much actual profit they will return; and iii) whether Tesla has the capital resources necessary to pursue it to the end.

You shouldn’t dismiss them if you want to actually understand the skeptics’ argument. The pivot seeks to avoid those hurdles, but these are hurdles that affect Tesla’s ability to execute the pivot. The pivot requires massive amounts of capital - far more than Tesla has on hand. Tesla isn’t a company like Google or Microsoft, whose existing businesses generate more than $100 billion per year in net earnings with which to finance moonshot ventures. Tesla’s net earnings are an order of magnitude lower and have been shrinking. Tesla’s earnings from automotive matter a lot to Tesla’s ability to get to the finish line - or how much of those possible future returns Tesla’s existing shareholders will get.

Not what it said. Musk may be the world’s most successful industrialist, but some of his projects fail. Solar roof failed. Cybertruck failed. 50% CAGR failed. Heavy automation of the auto factories (remember the dreadnaughts?) failed. Boring company has basically failed (it hasn’t materially lowered tunneling costs). Etc.

One of Musk’s main skills is being able to fail in many individual pursuits and to be successful overall - he’s not risk averse, he tries more ventures than will succeed so he doesn’t miss out on great opportunities, and he’s able to learn from his failures.

Just because Musk is a great industrialist doesn’t mean that he’ll be able to make AI that’s good enough to ‘drive’ a humanoid robot in any useful work. Just because Musk is a great industrialist doesn’t mean he’ll be able to create an AI driver that’s good enough to make a taxi fleet profitable. He might fail at those two ventures and still be a great industrialist on the whole.

You should be interested, because however you think intelligence works, Google’s free Gemini product is not an intelligence. It’s an AI tool, and it has certain attributes that need to be understood in order to use it properly. And one of those attributes is that it does what its designers tell it to do, which is to drive user engagement.

When you provide it an input, it is designed to provide you with an output that is likely to get you to want to keep using it. The most obvious example is that nearly every response ends with a few questions to try to solicit you to use the tool again. The less obvious one is that it will shade its substantive response to make you more pleased with what it has returned. It is designed so that you’ll like using it.

2 Likes
14

Or, you could go back to responding to people with an actual researched opinion (or at least going to the referenced articles in your AI to verify that the Gemini summary is even accurate (hint, it often isn’t), like we all did just a year or two ago before Gemini existed.

4 Likes
15

There are some bosses who only want to hear what they want to hear. Gemini is perfect for them, although I think they should have named it “Natalie”.

1 Like
16

I am finding more and more instances where the summary of Gemini is not supported in any way by the text of the source article. It happened again to day - and my perception is that the incident rate of such is greater than 50%.

Hawkwin

Who recently used ChatGPT over the last two weeks to research and purchase an additional 1TB internal hard drive and had to tell ChatGPT it was wrong about either the quoted price or the quoted TB (often giving me 500GB results) of a drive over a dozen times.

17

Again - TMF just needs to ban AI generated gibberish. If anyone ever wondered why, this thread shows exactly how worthless it can be. Thank you Hawkwin, Goofyhoofy, albaby, et al for the rebuttals that captainccs could not refute.

2 Likes
18

Indeed. AI models are actually not all that great for finding out current things about the real world. They don’t necessarily work the way you think they would.

I’ve encountered similar problems trying to use both Gemini and Claude to research things like restaurants or cars (when we were shopping for one). When I kept running into significant errors and omissions - especially big discrepancies between what they were telling me in a narrative and what was in the links - I got pretty frustrated and really dug into asking it why they were making these errors.

So, these models basically have a set of core parametric knowledge, which relies on the massive training data sets they were created on. For Gemini, that core parametric knowledge runs through about January 2025. For anything after that date, the AI has a search function - it can submit queries to a web browser tool. Unsurprisingly, for Gemini that web browser tool is basically google.

However, the AI’s use of the web browser tool apparently is very constrained. It won’t trigger the web browser unless the prompt clearly requires it - which one scenario that leads to a lot of hallucinations (the AI thinks it can answer the question with the parametric knowledge, but it’s wrong). Second, it won’t look into the web pages themselves unless the prompt clearly needs that, and instead will try to answer the prompt based on the snippets first. Third, it can’t actually see the web page the way we see a webpage - it the browsing tool only parses the rendered text and HTML structure of the page, and it can’t “see” images or anything else on that page. And finally, the AI is really limited in how much it can use that browsing tool - when looking for restaurants, I had to carve up my search into literally a dozen separate prompts, because the AI was only “allowed” to look up a few dozen web pages at a time.

In short, you can conceive of one of these free AI models as if it were an entity that came into existence with nearly all of the internet-based knowledge that exists as of about two years ago, sitting in a room with very limited Google access - and who will do everything it can to first avoid using the Google, then to just use the snippets to avoid clicking on any actual links in the Google, and then finally to click on no more than a few dozen links before giving you an answer anyway.

So, what happens when you ask it things about products on the web - or restaurants or other things like that? Free AI models don’t think. They don’t understand what you’re asking, and they don’t understand their output. What they’re doing is statistically generated the most likely string of characters that will be responsive to the inquiry.

When asked for something like this, the model often doesn’t “look up” an actual price quote. It generates text that looks like a real price quote because its training data is full of instances following that exact visual pattern. Because the model doesn’t “know” it is guessing, it generates fabricated quotes (or URL’s or restaurant information) with the exact same confident tone it uses for real facts. It doesn’t have a truth module, or any internal process that it can use to check whether its guess is correct or not - and it is heavily biased to always respond rather than say “I don’t know.”

You can try to hack around this with very detailed prompts that instruct the AI to ignore its parametric memory, rely only on information obtained from real-time web searches, to confirm every response with a live web search from an actual reviewed URL, to specifically return “no response” or something similar if it doesn’t do that confirmation, to direct it not just to return a response but explain how it got the information, etc. But even then, you can’t necessarily get around this. You don’t control the model, and you can’t actually give it instructions. You’re only giving it prompts, and instructions about how to do something are just another prompt. The model reacts to your instructional prompts the way it reacts to everything else - it generates a response that is statistically likely to generate enough of a positive response from you to elicit further interaction. Following your “instructions” is part of that statistical likelihood, but it’s not a constraint - and you can’t use prompts to change how it responds to your prompts, just to modify the prompt.

2 Likes
19

BTW, I asked Gemini to run the AI query that was the “conversation with Google AI” at the top of the page to get its response - and then I also asked it to tell me what it looked at to generate its response.

Gemini said it relied mostly on its parametric data, nearly a year old. The only recent data it obtained was to do two web searches:

  1. "investors Tesla pivot robots cybercabs humanoid Optimus 2026"

  2. "Tesla investor reaction Cybercab Optimus autonomous pivot 2026"

…and then looked at only ten hits. Not ten whole articles, mind you - for this response, it didn’t get past the snippet level. It looked at only excerpts of those articles as delivered by the web browser, approximately 3,200 total words - about 300 words each. The sampling bias on this is going to be astronomical. Running 2 or 3 quick queries will only grab a snapshot of what mainstream financial media says on the topic, and then only looking at one or two paragraphs of each article? Baffling.

I have to admit, I knew that AI was super-limited in what it looked at when generating responses to questions about current things, but I had no idea it was that constrained. I would have thought it would do more than the human equivalent of 10-15 minutes of reading excerpts of a handful of Google hits before creating a response.

20

What I want to know is why Fools insist on 100% perfection from Elon Musk. I’ll just post AI’s conclusion

GoogleAI:

Ultimately, the argument is that treating every missed target as a fatal flaw is a foolish way to evaluate an innovator who is fundamentally shifting global industries.

Sound conclusion!

The Captain