They’re buying the customers and the brand.

When they could instead sign with the new hottest enterprise coding IDE (Claude, Codex, etc who are way more popular now). Maybe if it’s an acquihire, it’s the GTM/Sales that xAI is after?

You should learn about securitizations. It’s actually interesting. But people talk about it colloquially and so incorrectly that it’s mind dumbing.

Here’s a simplified example of how you can take something and turn it into a safe investment:

Suppose you have 10 loans and each has a 50% chance of default. Ignore coupon, and say they are $10 each. Expected value is $50

If you were to put this in a deal and cut it up into tranches, say the first tranche gets the first $10, this would be your AAA bond because odds of getting paid out you $10 would be > 99.9%. The equity (bottom tranches) would pay a lot less. For instance the expected value of the bottom half would be considerably less than $50 that is being promised. So there’s upside since you’ll be paying cents on the dollar and even though in the median scenario you’re making nothing, you have to weight the expected values of each scenario to figure out how to price it.

The problem w this model is that it only works if assets are relatively uncorrelated which wasn’t true (it was true in the past but ignored systematic risk and adverse selection in originations).

What this has to do w musk or spacex I’m still not sure

Now why is this bad? Well, if you invest in a fund that is based off of the indices, you’re going to be investing in SpaceX whether you want to or not and I certainly don’t think 15 days is enough time to sus out whether this is a stable investment worthy of being in the index, but it’ll be great…until it drags a million retirement funds down with it.

Google and others were sitting at the corner, laughing that they gonna burn their money for no reason! they turned out to be wrong.

Turns out offering discounted/subsized tokens to developers massively improves your AI compared to just being a talking parrot for normal user workflow where you do not get "instant feedback" on if it worked or not.

Honestly, it felt much stranger to me to learn, a few years ago, that they're 3D-printing rocket engines. With my experience limited to building my own PLA/ABS 3D printer out of salvaged motors and parts printed on another printer, it was hard to imagine how this is anywhere near safe and precise enough. But turns out, FDM-ing some plastic blobs is not the same as fusing Inconel powder with lasers. Same with using LLMs for software engineering (whether in aerospace culture or otherwise), it's just not the same as asking ChatGPT "please make me an app to do something idk how i cannot code send halp".

They need xAI as a reason for the narrative that data centers will be in space, so SpaceX can project far more growth before the IPO. After the IPO they'll find out that data centers in space are too expensive and overheat.

The problem is they can't compete with Anthropic and OpenAI because they can't sell Opus and GPT at a discount to subscribers like OpenAI and Anthropic do with their subscriptions.

So they either need to build a competing model or slowly die.

for contrast, Elon paid $44b for twitter back in 2022. When you adjust for inflation, Twitter was acquired for $49b in 2026 money. Cursor getting bought for 1.22x more is just insanity.

Elon seems unwilling to shake off the image that he has basically no idea what he's doing.

More accurately it is 3.4% of SpaceX at the last rumored valuation of $1.75T.

Note that Meta paid ~16b for Alexandr Wang, and Google paid ~3b for the windsurf executive team. You are making a category error -- the talent pool isn't "ML researcher" it's "competent leader"

I see two possibilities:

(1) SpaceX is paying with stock; and

(2) the $60bn pay-out is (a) conditional or (b) never going to be exercised—it was a stalking horse for negotiating the $10bn terms, which gives SpaceX everything it actually wants.

What you've described is how the base level mortgage-backed securities (MBSs) work. The tranches work because there actually exist mortgages that are at lower default risk (high home equity, well qualified borrowers, etc.), and the senior tranches are effective in capturing their underlying safety. What CDOs did was to take the lower, riskier tranches of MBSs from various sources and repackage them and divided them into tranches again. Then they got the ratings agencies to rate the top tranches of the CDOs as AAA as well. It's as if a teacher graded several classes and then took everyone that got a C or below from all the classes and then graded them on a curve again. And suddenly a lot of the C students became A students. It was outright financial insanity. Well, mixing a rocket/satellite company with a couple of also-ran AI outfits and the walking corpse of Twitter, and then calling the whole thing SpaceX and valued at $1.75T is a similarly level of financial insanity to me.

And that naive statistical reasoning is where it goes terribly wrong. You have to consider the causal process that generates that distribution!

The type of people who would default on a coinflip are extremely sensitive to how the economy changes. The probabilities are very correlated, the expected value is rather meaningless then. It's closer to having a 50% chance to either get a full return or get zero returns, depending the macroeconomy, quite the gamble. Actually, those people were in a rather dodgy situation in the first place, or are not great at decision-making, so it might be more like 50% chance either of getting 50% return or getting 0% return.

PS: Just elaborating on your point, not meant as a counterargument, I know you said the same thing.

I think you meant "the chances of getting paid", not of not getting paid.

What this has to do with with SpaceX is that there's the same blatant disregard for sound financial analysis by the very institutions that were/are supposed to know better. The NASDAQ 100 fast track decision is a similar level of financial malpractice as the ratings agencies slapping AAA on things that they knew were little better than junk. The abuses of the subprime mortgage originators were well known long before the actual meltdown. As were those systemic risks you spoke of. They were ignored by those whose entire job it was to not ignore them, and they sold out their credibility for a quick buck. If you can't see the similarities to the present situation then I can only wish you luck.

Well, I said no. Not getting burned that way again!

And xAI now gets 10B of more revenue on their income statement.

Perfect financial statement boosting for the IPO which in turn will pay back these costs.

At least that’s the bet.

A controlled environment to determine effort and token usage, and to get plenty of exclusive training on code.

It could end up making sense. Idk if they needed to offer 60B though.

I think Gemini is best model out there, and it's not Cursor who you should praise. I use it with jetbrains junie. Vastly cheaper than claude, faster, produces better quality code, actually listens to your instructions, more accurate. I'm sure claude code cli has some cli magic that I'm missing out on, but having everything just work in a nice IDE (and llm to actually understand your symbol table) is like magic.

it might not help all that much once it turns into "grok" harness or otherwise associated with elon

Anthropic just tolerates the money losing developers who pay $20/$200 for subscriptions.

Circa 2003

Composer-2 seems very useful in Cursor, while K2.6 according to AA seems to be a really useful general model: https://artificialanalysis.ai/articles/kimi-k2-6-the-new-lea...

Has that changed now?

Thanks!

When coding agents work they're great. When they don't I still need the IDE. They usually don't work that great when I'm working on something novel or brownfield. Which happens quite regularly.

But I definitely still want ai autocomplete. I'm not a Vim user. Coding isn't about typing for me, it's about solving problems. So a tool that does lots of the typing for me is a godsend.

So do I go for VS Code + Copilot? Because it was bad when I tried it again for a few days in November. Slow to respond and gave poor results. Cursor is snappy and gives useful results most of the time.

They are catching up fast!

https://www.businessinsider.com/chamath-palihapitiya-ai-cost...

If you think this of users who use cursor then I don’t think you’ve used cursor much at all.

Also one would definitely offer to pay in stock if they believe it is massively over-valued lmao.

Sometimes it helps to go back to the basics to understand company performance: money in, money out?

Mortgages are very cuspy. It's pretty wild that someone would give you a 30 year loan with 20% equity for a few percent higher than risk free. Also you could default on that loan and they can't garnish your wages. And if you default, your credit history would reset after 7 years. Oh and you can repay the loan at no cost, so if rates go down you can just pay it back and turn around and get another loan at a lower rate, or if rates go up you can hold on to it until 30 years.

It's the same thing with CDOs. You take something that has some undesirable characteristics (these cuspy BBB), structure it in such a way to create some safe and riskier assets. And hopefully the sum of the final tranches is worth more than the components.

It's like if you were forced to sell an animal whole. The individual components are worth more because people have different preferences. With CDOs (excluding synthetic), the amount of exposure is unchanged. It's a bit more concentrated where the riskiest parts are in these CDOs, but nothing changes.

I get that finance isn't really sexy and people see it as just pushing paper around, not creating any value. But there's real value in taking some components and creating something more valuable with it. It's like using flour + sugar + egg to create cookies worth a lot more than the individual components. There was fraud and negligence but people are mad at the wrong things.

Rating agencies did a poor job, but in their defense, delinquencies and defaults reached levels well outside expected values due to systematic risks. Also rating agencies are kind of a joke. Investors aren't dumb. Even today, look at debt, there's a big difference between bonds of the same rating and similar weighted average life.

The bad thing about rating agencies is how regulations rely on them to determine what "safe" is and capital requirements. Of course, mandated capital requirements shouldn't be the end all be all of risk management, but these guidelines that over rely on rating agencies don't help the matter.

Mixing rocket company with AI and social media is fine. It's just a conglomerate. Who cares? Look at Samsung, they sell smartphones, TVs, ships, they're involved in construction, even insurance and biotech.

The question is what is the underlying core competency they're relying on and it's obviously Musk. And he has been able to deliver innovative products (manufacturing and forward thinking technologies). He scaled up one of the largest training clusters in the world in a very short period of time. He created a large car company after decades of stagnation. He lowered cost of getting stuff to orbit by orders of magnitude and now handles something like 90% of rocket launches. He's gotta be doing something, right?

I do agree that the optics of this aren’t great, and it’s rather easy to be cynical about motives.

I'm not an expert but it looks to my like 80% of my allocation won't be tracking spacex, because it's mid cap or small cap etc, and the 20% that's in the vanguard growth index might? I assume whoever sets the rules for the fund could change the rules to say companies must be listed for X months if they want to avoid this, right?

And I can change my allocation.

edit: Actually wait, isn't it only nasdaq 100 that's tracking it early, after 15 days rather than 3 months of trading? So 0% of my 401k is exposed to buying it quickly after IPO already, I think.

If so that would seem like the most plausible take on why this is happening.

Cursor is essentially all the Claude Code products but without the horrible bugs of Claude Code products.

You can transfer from CLI to web and it actually works.

That said, people are increasingly migrating to CLI tools (Claude Code if you like the Claude models, Pi Agent if you want something that's highly customizable, Crush if you want something fun), or GUI tools that are less code-first (Codex GUI).

That' all well and good and they had astounding growth rates but doesn't mean much. And 1B in ARR is not _that_ much in comparison. Also, reportedly they spend all their revenue and they have no control over the spend-side. The models they use will very likely get much more expensive. All the foundation model companies have a competing product. Cursor has the first mover advantage, but that will only help then so much. There have been plenty companies who grew fast, had huge revenue, but failed in the end, because they never got profitable. That's also in the cards for Cursor, if they don't fundamentally change their business model

SpaceX has invested a small amount as a share of its value in XAI, and could survive the loss of its investment.

Wild conjecture.

1) A gamble based on cursor's compute constraint 2) if 1) plays out, he can purchase cursor via shares of spaceX over valued shares, at a fixed price should the valuation increase.

The main frenzy with Cursor started when you could access Anthropic models practically for free.

Otherwise it is just VS Code.

And on the AI development side they're the ones providing compute in the form of a "million H100 equivalent Colossus training supercomputer"... On top of the cash.

It is true that they were not transparent about the base model that they used until the model slug was discovered by a Twitter user via the API.

[1]: https://arxiv.org/abs/2603.24477 [2]: https://cursor.com/blog/real-time-rl-for-composer

https://web.archive.org/web/20150822195811/twitter.com/valen...

https://twitter.com/valentine_

(If any lawyers read this and feel up for taking this on contingency, I don't think I'm difficult to contact.)

That being said, I noticed that Kimi being served through Openrouter providers was trash. Whatever they do on the backend to optimize for throughput really compromised the intelligence of the model. You have to work with Kimi directly if you want the best results, and that's also probably why they released a test suite to verify the intelligence of their new models.

I found it much more consistent than glm or minimax

I guess really depends on tastes

But you can't actually sign up to Pro or Pro+; they disabled sign ups until the per token pricing starts.

I guess I don't really understand what it buys you over just running vanilla VS Code and Codex.

https://x.com/chamath/status/2029634071966666964

I suspect that as the value a company provides is more than its code, then increasing code churn does not lead to an equivalent increase in revenue. Even for a tech company, a business' concept, connections, knowledge, assets, non-coding staff, etc.. are a significant value and increasing code doesn't increase the throughput of that value. For non-tech companies code is the grease in the gears, not the gears themselves.

Only the foundation model companies offer cheap/subsidized compute.

If you're an app layer company, you're offering a 10x worse deal to your customers.

Foundation model companies are willing to lose money to win loyalty. Remains to be seen if it'll work.

They removed it largely because investors wanted higher returns, and the tech companies that had such dual classes (1) were doing really well, and the S&P ended up caving on that rule.

1: Perennial hot button around here Palantir did this in a more extreme fashion than most. The three founders F class shares will always be at 49.9999% of the votes and the early investors B class shares have 10 votes each as compared to the publicly traded A class shares 1 votes.

S&P 500 includes companies from multiple exchanges. Like Nvidia, which lists on Nasdaq.

true, but its not worth $60 billion fucking quid.

This. They are after the harness engineering experience of the Cursor people, I'd assume the they want to absorb all that into Grok's offerings.

The value and the room for innovation on the harness side seems to be underestimated.

Oddly the harness also affects model training, since even GLM/Z.ai for example train (I suspect) their model on the actual Claude Code harness. So the choises made by harness engineers affects the model. For Kimi/Moonshot and OpenAI the company makes their own harness. Alibaba uses Gemini.

Very interesting dynamics.

The way I work now in the Codex desktop app is that I spin up 3-5 conversations which work in their dedicated git worktree.

So while the agent works and runs the test suite I can come back to other conversations to address blockers or do verification.

Important is that I can see which conversation has an update and getting desktop notifications.

Maybe I could set this up with tabs in the Terminal, but it does not sound like the best UX.

  you've formed an opinion on the value of the company without knowing how many users it has? Kind of proves my point, no?

This is a bit simplistic. It's the VS Code that everyone used before cc came to town. Real devs, on real projects. All that data they collected is worth a lot more than "just vscode". Their composer2 is better than kimi2.5 and it's just a finetune on that data.

xAI had a decent model in grok4 (it was even sota on a bunch of benchmarks for a few weeks), but they didn't have great coding models (code-fast was ok-ish but nothing to write home about, certainly nowhere near SotA). Now that they've been banned from using claude, they'll get their expertise + data to build a coding model on top of whatever grok5 will be + their cluster for compute.

It doesn't sound like a bad plan to me, financial shenanigans or not.

Cursor needs their own 1st party backend model.

Sounds like a match made in heaven.

But I agree that it's hard to articulate what Cursor services you could blow this much money on.

Maybe it is all just an option! Or maybe they get a bunch of IP either way?

I didn't say it was Wise.

I said it seems within possibility for this, very particular, corporation.

e.g. https://arstechnica.com/tech-policy/2022/12/twitter-stiffs-s...

I wouldn't trust a contract from one of Elon's companies unless they were willing to put in escrow an amount that would make me whole in case of a breach on their side. (And that amount would be quite large in the case of a potential breach involving using prompt data for training.)

Best I can do is pretend to be a lawyer and forward all of ur stuff to ChatGPT Free. U down?

This is digital feudalism, and the billionaires have seized the means of communication.

The thing is, every dollar you spend on insurance is a dollar (and its interest) you lose. Furthermore, we don't know when it will pop. 1 year? 5 years?

The more reasonable solution is probably gradually reduce exposure to US markets by selling SP500 shares and turning to Europe and emerging markets ETFs. No need to cash out 401k.

https://www.morningstar.com/funds/spacex-ipo-how-index-funds...

> Nasdaq was the first to consider a rule change that would grant mega IPOs like SpaceX early admission to its flagship Nasdaq-100 index. The exchange and index provider began a consultation period in February to assess the viability of and industry response to a proposed “fast entry” rule. The change was approved on March 30 and will be effective on May 1.

EDIT: Parent commenter edited out the @dang from their comment making mine appear to be responding to something not in their comment.

We're probably looking at nuclear fission generators to get started, then converting to geothermal at any appreciable (and maybe fusion, inshallah).

Cursor's token utilization is significantly better than Claude Code. Composer's latest model, for coding, is very competitive on quality given price and was clearly well-optmiized (in two months, you will hear almost nothing else than how expensive Anthropic is...this is before they try to release the really expensive models). so many very obvious things like this if you have been using this tech every day for multiple years.

unfortunately, the competition in this space is very weak because of how dominant cursor has been (Kilo/Roo/Cline all have major implementation issues with token utilization, everyone else is trying to go all in on agentic). don't see this getting better until things get much worse because of anthropic/agentic. from the decisions that anthropic is making, it seems they are busily digging their own grave. growth will come after this.

Be it the Dyson shell thing or Lunar or Mars colonies, there's no way it'll be done relying on transports from Earth surface. It could only work if we could make them from asteroid pieces. Which makes most items on their tech tree from Starship forward obsolete. And they're already all-in on those techs. It makes so little sense in so many levels.

My point is regardless of what you think of a Dyson sphere, this theory seems to predict what the company does better than assuming everything's a ketamine fever dream.

https://youtu.be/hjdMYyjnmks?si=iyoVV-oZAPmQtp1B

Sure. My question was why. And my loose interrogation of the question, together with some unique domain expertise, suggests he found an excuse to work towards a Dyson sphere.

Though, in fairness, that's probably the important part. Like a base model plus "coding smarts" is probably perfect for the situation.

But maybe not as much value as I was thinking.

My account was hijacked via domain/DNS takeover around the time it was acquired by fElon (due to both Crazy Domains and Twitter support's incompetence — both parties removed 2FA from my accounts, even despite me telling Crazy Domains specifically never to do so). I managed to recover both accounts after kicking up a fuss, but the hijacker was midway through an 3rd party account wiping script, and I'd lost all my followers because of that.

I had 33,300+ tweets in 2015, and a lot of that was private interaction with friends.

Digital decentralized protocols (smtp, http, etc) were the first time this wasn’t true. But you [we] voluntarily moved your communication off of open internet protocols onto private ad-based platforms.

Of course you don’t own anything there, you never did. The billionaires didn’t “seize” anything. You happily sold yourself out for a few clicks of less friction and an easier shot at digital fame by going “viral” on social media company land.

If this isn’t a much bigger indictment of the collective (who after decades still could not agree on a non-elitist, human understandable protocol that didn’t require a CS degree to use) than it is of the entrepreneurs who solved all the problems the collective refused to, I don’t know how else to get though to you.

You can do that with claude code, github copilot (built into vs code) and codex, in any of their IDE versions, plugins for other ides (jetbrains, vscode, anything else you care to name) and also, of course, the CLI versions of all of them. They're also integrated into github, jira, and everything else.

Seriously, try other tools! if only to get a more balanced perspective.

This all being said, its been a long time since I last tried cursor... I'll give it a go.

I'm trying to help my parents now their at retirement age and am seeing first hand what not planning for your future looks like. They hit retirement with nothing but a small social security check every month. Not even enough to cover rent in most places.

I don't know how much you have in your 401k, but it will be worth literally hundreds of thousands more if you pull it out when you retire. You aren't just paying the penalties now, you're paying for potentially decades of compounding.

I'm not saying they aren't profitable. I don't know, but it's definitely not a given.

I think you're going to be surprised at the level of competition BO provides SpaceX in the Artemis program.

Did you notice the size of the Artemis rocket and the size of the payload it sends to the moon and back?

Do you expect there to be diamonds just laying these on the moon surface, no mining required.

What I see a lot of people do, unfortunately, is reconcile this contradiction by not following the published limitations of the "Full Self-Driving (Supervised)" product. They assume that Elon Musk wouldn't call it that if it couldn't be trusted to do what they expect. Then they get into fatal crashes, and someone sues, and Tesla argues that they can't be held accountable for bad drivers who don't follow the rules.

But, even if you want a big all-in-one editor in an Electron app, it seems obvious VS Code is the way to go (or Zed, if you you aren't committed to using an Electron app). I just can't think of anything Cursor offers that makes it worth spending extra money for it.

A team could build an AI IDE in a week, this could be a race to the bottom

I think Musk being a ketamine addict explains a lot and this is very aligned. Given that he can't build a self-driving car, he would have to be under the influence of very strong drugs to think he could build a Dyson sphere of all things.

That being said, some sort of financial fraud is even more on brand.

Unsure how it would work in practice.

> If you enable “Privacy Mode” in Cursor’s settings: zero data retention will be enabled for our model providers. Cursor may store some code data to provide extra features. None of your code will ever be trained on by us or any third-party.

Note the "may store some code data" and "none of your code will ever be trained on". In general you never want to include actual customer code in training the data, because of leaks that you may not want. Say someone has a hash somewhere, and your model autocompletes that hash. Bad. But that's not to say you couldn't train a reward model on pairs of prompts + completions. You have "some code data" (which could be acceptance rate) and use that. You just need to store the acceptance rate. And later, when you train new models, you check against that reward model. Does my new model reply close enough to score higher? If so, you're going in the right direction.

> If you choose to turn off “Privacy Mode”: we may use and store codebase data, prompts, editor actions, code snippets, and other code data and actions to improve our AI features and train our models.

Self explainatory.

> Even if you use your API key, your requests will still go through our backend!

They are collecting data even if you BYOK.

> If you choose to index your codebase, Cursor will upload your codebase in small chunks to our server to compute embeddings, but all plaintext code for computing embeddings ceases to exist after the life of the request. The embeddings and metadata about your codebase (hashes, file names) may be stored in our database.

They don't store (nor need to store) plain text, but they may store embeddings and metadata. Again, you can use those to train other things, not necessarily models. You can use metadata to check if you're going in the right direction.

So no it’s not an oxymoron.

If you just look at the past 20 years, the US has had exceptional returns compared to the rest of the world.

The thing is, historically, high PE ratios like what we're seeing in the US do not correlate with short term returns that are as high. Expected future returns decrease as the PE ratios go up in a pretty linear fashion.

https://am.jpmorgan.com/us/en/asset-management/institutional...

If you want a different point to backtest from, try Japan in the 80s and early 90s

I don’t personally think Google is worth $4T but the share price says otherwise.

Again, not debating that SpaceX isn’t a legit company or that it’s profitable. But underwriters agreeing with high valuations to stocks that collapse once they go public isn’t unheard of.

Edit: and I will concede that I should’ve phrased my initial thoughts better. Credit rating agencies and underwriters do very separate things, just like IPOs and MBS are two very separate things.

Any argument about how people don't pay enough attention since it isn't yet certified as a L4 system is irrelevant and tangential to the point.

The characterisation of “level 5” autonomy as the car handling any conceivable circumstance (not that you explicitly made this claim here) is just silly. Humans can’t handle any conceivable circumstance either.

But if by some tragedy you don't die young, your older self is gonna be pissed at younger you for costing him hundreds of thousands of dollars.

Really? We're still making claims like this in the year of our Lord 2026? People in the markets today are not predicting the real value of a company, they're gambling that the various political and financial machinations from people like Elon Musk will increase the share price enough that they can sell at a profit. The value of shares like Tesla are utterly disconnected from the value of the underlying business.

That isn't what is happening at all.

In an IPO the underwriters and the company collaborate to set the price based on approximate demand and what they want the quality of the holders to look like.

In the roadshow, the company is very constrained as to what they can say or disclose outside of the scope of the S-1. They can't include MNPI, forward looking financial projections, etc. Underwriters are also prohibited from sharing MNPI, or publishing marketing disguised as research.

So I guess if you're saying the SpaceX S-1 is completely full of shit and there's hidden risk in it, than it could be similar to 2008, but in this case nobody is manufacturing a rating, and those material misrepresentations would constitute securities fraud. Investment banks and ratings agencies aren't the same thing at all, and the buyers of marginally profitable IPO stocks are (hopefully) different than those of AAA MBS.