M&A: xAI, A SpaceX Company

[Vultur]

Or, as it applies to your sloppy argument about AI:
On what timeframe is AGI achievable?
Are LLMs alone sufficient to reach AGI?
If you're going all in on AI, how should you go about it--Invest it all in chip design? fabrication? model building? datacenters? pure research?
What happens if your estimates are directionally right, but you misjudge the resources or time required to achieve them? Can your company survive an unexpected change in macroeconomic or market conditions?
If you achieve AGI, will it be a winner-take-all market, or will you still be in competition in the form of competing AGI's?

You can be fully committed to the concept of AI and believe it to have infinite upside, and still come to very different conclusions about how to act based on how you answer those questions.

Essentially, leaving the religious side out of it, there is the possibility that the current investment is not actually progress toward AGI* in the sense that further investment in the same direction will get us there.

In one sense, inventing balloons got us "closer" to space than climbing mountains did. But it required a fundamentally different technology (rockets) to get all the way there.

*I am skeptical that this is terribly well defined, or that achieving it by some definitions would automatically have the overwhelmingly huge impact that seems generally assumed. But that's a different argument. Even granting both for the sake of argument, if the balloons vs rockets analogy holds, investment in the current models would have zero effect (or negative effect, since it removes money that could instead be used for something else) on the chance of getting there.

[SpaceLizard]

Worse than that--xAI not only hasn't been able to stand out in terms of enterprise marketshare, but the most notable times Grok has broken through to the public consciousness has been embarrassing failures like the MechaHiter incident, it's bizarrely effusive praise of Elon, and the recent sexual deepfake scandal.

None of these failures are unique to xAI. Microsoft's Sydney threatened users, Google's Gemini portrayed people of color as Nazis, both Google and OpenAI has sexual deepfake problem too, and currently there are several lawsuit against OpenAI due to ChatGPT's role in suicides.

AI is new tech, some misstep is to be expected, it's no different from the limited environmental impact Starship caused at Starbase. You can choose to singularly focused on the negative but the reality is the positive is far greater.

The fact that AI gives me correct information about 90% of the time means it has negative usefulness right now.  It takes me more time to figure out which 10% is wrong and get the correct information that it would take me to get the right information in the first place.  I've also had people I work with provide me with AI-generated text, which to this point has contained errors of fact 100% of the time.

Would you trust a calculator that gives you the correct answer 90% of the time?

It's not a calculator though... It's a glorified search engine and content generator with a veneer of personality...

[Coastal Ron]

...
I had FSD negotiate a multi-lane four-way intersection, stop lights were out, everyone was kinda haggling over who was there first.  The car handled it extremely well, asserted itself without being reckless, basically did what a human does.

That was during the demo a few months back, I don't pay for it since I like driving too much...

Which is why I have never heard anyone make a choice for buying a Tesla because of FSD.

The market for autonomous cars is likely to bifurcate between ~Level 2 assisted driving and Level 5 Full Automation. Essentially anything with a steering wheel for users will require user supervision and input (i.e. Level 2), where services such as taxis and shuttles will be fully autonomous (i.e. no user input).

And car autonomy doesn't require a car to be electric, which is the challenge for Tesla, especially during the current Trump Administration. Which is why Tesla cars sales are likely to continue to slump.

They'll go to L4 at some point - it seems very close to ready.

As I age, I will want safe autonomous forms for transportation that I'll be able to call on demand. Waymo seems to be the furthest along for doing that, not so much Tesla. Plus, the Tesla Robotaxi is the wrong form factor for older populations that need to make short trips. Tesla can correct that design choice, but that would be more $$$.

As for how soon, again, Musk said FSD was going to be available in 2018, and every year after that. Did Musk ever say WHY he was so wrong? Because unless someone can understand and explain why their public declarations are so wrong, I would think it would be prudent not to believe everything they say. Right?

And just to clarify, I'm not saying "never" for any of these topics, my whole focus is "when". Musk seems to eventually get stuff done, but hardly ever on the timeline he announces.

But unlike with Tesla and SpaceX, xAI has fierce and well funded competition, and tech companies are not waiting for "data centers in space" to build out more compute capacity.

[Vultur]

Worse than that--xAI not only hasn't been able to stand out in terms of enterprise marketshare, but the most notable times Grok has broken through to the public consciousness has been embarrassing failures like the MechaHiter incident, it's bizarrely effusive praise of Elon, and the recent sexual deepfake scandal.

None of these failures are unique to xAI. Microsoft's Sydney threatened users, Google's Gemini portrayed people of color as Nazis, both Google and OpenAI has sexual deepfake problem too, and currently there are several lawsuit against OpenAI due to ChatGPT's role in suicides.

AI is new tech, some misstep is to be expected, it's no different from the limited environmental impact Starship caused at Starbase. You can choose to singularly focused on the negative but the reality is the positive is far greater.

The fact that AI gives me correct information about 90% of the time means it has negative usefulness right now.  It takes me more time to figure out which 10% is wrong and get the correct information that it would take me to get the right information in the first place.  I've also had people I work with provide me with AI-generated text, which to this point has contained errors of fact 100% of the time.

Would you trust a calculator that gives you the correct answer 90% of the time?

It's not a calculator though... It's a glorified search engine and content generator with a veneer of personality...

What makes LLMs troublesome as a "glorified search engine" is that they report all their results, correct and wrong, in a consistent "LLM voice" and out of context.

Back when Google was good, you found both true and false information in searches.  But those were on different webpages written by different people, in different styles and contexts, so you could judge the trustworthiness of the source. An LLM removes those trustworthiness cues.

And just to clarify, I'm not saying "never" for any of these topics, my whole focus is "when". Musk seems to eventually get stuff done, but hardly ever on the timeline he announces.

But unlike with Tesla and SpaceX, xAI has fierce and well funded competition, and tech companies are not waiting for "data centers in space" to build out more compute capacity.

This is a key point. A year or two delay could make a huge difference here. Even on the ideal timeline, they're not launching these things this year.

In the next few years, a lot of relevant things could happen:

- someone (*cough* Google *cough*) could "lock in" much of the demand, so even a technologically superior competitor can't get enough market share to make a profit. (Windows hasn't been technically superior in ages but business, government, etc. keep using it.)

- The bubble could burst, and there could be massive excess data center capacity like there was massive excess fiber capacity in the early 2000s. Thus no market for building more, in space or on the ground.

- Even if there isn't a 2000-style burst but a less dramatic market correction (for example, because MS, Google, Meta, Amazon have way more revenue than the 2000-era companies did and so can accept large losses without disaster), building new data centers could still slow dramatically because the profit is just not there ... So same as above, no real demand for more data centers (Earth or space), just without a giant crash.

- There's also the government angle. There's a possibility that an AI-unfriendly administration and perhaps Congress takes over at the next election, and puts in restrictive regulations and perhaps laws.


--

It's also possible that Starship will 'work' but not be cheap enough, because of TPS limitations limiting the number of reuses or requiring large amounts of labor per reuse.

[Coastal Ron]

This was something I thought needed to be clarified:

But Elon Musk can't force customers to choose xAI over all the other alternatives, and some of those alternatives have structural advantages that xAI (now SpaceX) doesn't, such as existing customer bases where adding in-house AI to everything means that users kind of get locked in.

That isn't everything (Microsoft is clearly struggling to get people to use their AI solution despite massive captured market of Windows PCs, see my link where they weren't even mentioned.) Also you are straight up ignoring the whole x.com user base. Not as huge as some others, but contrary to many claims in the media it did not die and has expanded in use.

X.com doesn't make money by offering AI services, it makes money by selling advertising. Users posts are inputs to xAI (which is scary if you think about it), but users don't post on X.com because of xAI - it is a social media platform that is free to use in its basic form.

Other AI companies, like OpenAI and Anthropic, make money from AI services, such as subscriptions and API usage. Google Gemini is being integrated into everything Google does, so it is a "sticky" engine of growth for them.

So X.com is a source of input for xAI, but from what I can see not a source of direct revenue.

[meekGee]

It's not a calculator though... It's a glorified search engine and content generator with a veneer of personality...

It is absolutely not a calculator and never attempted to be one.

But it is an entirely different beast from a glorified search engine.

The ability to pull information from "all of the Internet" is the LEAST interesting feature about it.  The ability to reason and keep track of a complex conversation involving many objects, subjects, concepts, and still be able to interpret "it" correctly even when semantically it's ambiguous (like <---) and to really understand the conversation well enough to continue a back-and-forth using very sparse language, that's the amazing part.

That it knows enough about me to make correct assumptions about things we discuss, and if the assumption is wrong, be able to take input and correct course - that's not found in any corner of the internet.

So I don't know which mechanisms are responsible for that behavior, but it's light-years past "search engine", even now.

[CoolScience]

It seems you gave up defending that junk article you posted should I take that as a concession that it was inaccurate?

Here is the first thing I found when looking up market share:
https://biggo.com/news/202602061122_ChatGPT_Market_Share_Slips_as_Rivals_Gain

Junk article? Not sure if you mean the Associated Press article, or the one from Menlo Ventures. But let's compare sources of information here:

The context should have been quite clear, the statement you quoted there was about the article you had stopped defending. That one has very clearly been debunked.

Menlo Ventures is a prominent, long-standing Silicon Valley venture capital firm with over $6.8 billion in assets under management (AUM) as of late 2025/early 2026.

Absolutely ludicrous, let's hear about the trustworthiness of VC according to your own words:

The venture capital market is famous for being lemmings, so explain what you mean with examples that are not related to VC investments.

Your Source: BigGo, based in Taiwan, and their tagline on Crunchbase is "Save more at BigGo, One stop service to find products, compare price, check price history with deals & promotion from all Online Shopping."

Yet another lie. That is the link I quoted, but not the ultimate source, the data comes from "Apptopia data via Big Technology."

Here is the link to the next layer down in the onion of sourcing, not that it actually matters.
https://www.bigtechnology.com/p/new-data-openais-lead-is-contracting

My point was simply that none of the metrics are reliable too many ways to add caveats to make it say whatever you want in such a new and changing market, so if you have some criticism of the actual original source, you will still just be aiding my point.

That you criticize without even properly reading the source and simply ignore many counterarguments and the actual points being made speaks volumes about the validity of your arguments.

[Vultur]

It's not a calculator though... It's a glorified search engine and content generator with a veneer of personality...

It is absolutely not a calculator and never attempted to be one.

But it is an entirely different beast from a glorified search engine.

The ability to pull information from "all of the Internet" is the LEAST interesting feature about it.  The ability to reason and keep track of a complex conversation involving many objects, subjects, concepts, and still be able to interpret "it" correctly even when semantically it's ambiguous (like <---) and to really understand the conversation well enough to continue a back-and-forth using very sparse language, that's the amazing part.

That it knows enough about me to make correct assumptions about things we discuss, and if the assumption is wrong, be able to take input and correct course - that's not found in any corner of the internet.

So I don't know which mechanisms are responsible for that behavior, but it's light-years past "search engine", even now.

The thing is, though, holding conversations isn't itself a source of much revenue. The glorified search engine thing is something an enormous number of people currently use AI for, and Google has demonstrated that you can monetize search very effectively.

AI as chatty companion is cool, but it's far from clear that it's a meaningful slice of the multi-hundred-billion-dollar-a-year market they'd need to make a profit on a several $T hardware investment with a relatively short depreciation time.

[Lee Jay]

Worse than that--xAI not only hasn't been able to stand out in terms of enterprise marketshare, but the most notable times Grok has broken through to the public consciousness has been embarrassing failures like the MechaHiter incident, it's bizarrely effusive praise of Elon, and the recent sexual deepfake scandal.

None of these failures are unique to xAI. Microsoft's Sydney threatened users, Google's Gemini portrayed people of color as Nazis, both Google and OpenAI has sexual deepfake problem too, and currently there are several lawsuit against OpenAI due to ChatGPT's role in suicides.

AI is new tech, some misstep is to be expected, it's no different from the limited environmental impact Starship caused at Starbase. You can choose to singularly focused on the negative but the reality is the positive is far greater.

The fact that AI gives me correct information about 90% of the time means it has negative usefulness right now.  It takes me more time to figure out which 10% is wrong and get the correct information that it would take me to get the right information in the first place.  I've also had people I work with provide me with AI-generated text, which to this point has contained errors of fact 100% of the time.

Would you trust a calculator that gives you the correct answer 90% of the time?

It's not a calculator though... It's a glorified search engine and content generator with a veneer of personality...

That was an analogy.  The point is, if I ask it a question and it gives me a wrong or misleading answer 10% of the time, it's worse than useless.  And it's not that good yet.

[Lee Jay]

It's not a calculator though... It's a glorified search engine and content generator with a veneer of personality...

It is absolutely not a calculator and never attempted to be one.

But it is an entirely different beast from a glorified search engine.

The ability to pull information from "all of the Internet" is the LEAST interesting feature about it.  The ability to reason

They cannot "reason" at all.

and keep track of a complex conversation involving many objects, subjects, concepts, and still be able to interpret "it" correctly even when semantically it's ambiguous (like <---) and to really understand the conversation well enough to continue a back-and-forth using very sparse language, that's the amazing part.

That it knows enough about me to make correct assumptions about things we discuss, and if the assumption is wrong, be able to take input and correct course - that's not found in any corner of the internet.

So I don't know which mechanisms are responsible for that behavior, but it's light-years past "search engine", even now.

The trouble is, it's confidently incorrect.  I've used them a lot, and I've so far gotten exactly one entirely correct response, out of perhaps a couple thousand tries.  Most of the responses are at least partially correct, some a mostly correct, but only one was fully free of errors of fact.

You don't see an issue with that?  The AI industry sure does.

[spacenut]

As noted, their car sales may be down, but their energy storage sales are up.  They do install industrial size storage battery stations as well as Powerwalls for homes.  Cars are not the only thing Tesla produced.  Also, some car companies increased in sales, while others were below Tesla.  I think Tesla should make hybrid cars that would be more appealing to those who still like gasoline vehicles.  I also think again, that they should rethink the bed situation for the Cyber truck to standardize the bed to accommodate standard tool boxes, pipe racks, camper shells, etc.  They could then push the sales to electric utilities, electricians, plumbers, and others who buy trucks to work out off.  It would be a great around town work truck with low range that could be recharged every night to work all day the next day.  Hope Elon Musk reads this.  Cyber truck was/is too much si-fi and not enough pragmatism for today's world. 

That being said, the in space data centers if shared could be a boon for others to avoid the pitfalls of more power consumption.  Right now the quickest way to get extra power is natural gas, not solar, not wind, not nuclear.  Gas generators are quick and easy to install and cheaper than the initial cost of alternative power sources.  If AI grows too fast, the power grid can't handle it, thus leaving space the only real alternative.  Solar has to be coupled with batteries making it expensive, windmills are location sensitive, nuclear takes about 10 years to jump through all the hurtles, thus leaving only natural gas as the quickest alternative.  Also, natural gas is more expensive than coal so there is another problem, monthly bill paying for power.  Once deployed in space, no monthly bills for power anyway. 

Musk is not the only person saying data centers could be in space.  Others have also said this.   

[Chris Bergin]

This thread continues to devolve into things that have nothing to do with this site, even with a trim about religion. Maybe take it to reddit as we have rockets to discuss here