On the eve of SpaceX's listing, what I am most curious about is the fate of Grok

SpaceX is about to officially offer shares to the public on June 11 and list on the stock market on June 12, raising no less than $75 billion, with a valuation of $1.77 trillion. There is already a lot of information online about the details of this largest IPO in human history. The prospectus and roadshow PPT are also publicly available, so I won't go into details here. SpaceX divides its business into three major segments:

• Space: Mainly space launch services, including the reusable spacecraft - Starship, which will start commercial operations in the second half of this year (previously declared "unrealistic" by many experts).
• Connectivity: Primarily satellite communication services represented by Starlink. This business was once used by communication industry expert Lü Tingjie (who often gives speeches at brokerage strategy meetings) as evidence that "Musk doesn't understand communication."
• AI: Mainly the former xAI business, including data centers, Grok, X (formerly Twitter), and the chip factory jointly built with Tesla, etc.

First, Musk merged X Platforms into xAI, and then merged xAI into SpaceX. From the perspective of capital operation, it couldn't be more brilliant. In the first merger, the users of X and Grok can be combined for calculation, resulting in the data that "xAI has a monthly active user base of up to 550 million." In the second merger, not only did xAI enter the capital market in advance, but SpaceX also got a bigger story. In the roadshow PPT, the management emphasized that the market size of the space and connectivity businesses is $2 trillion. The potential market size of the AI business is as high as $26.5 trillion, accounting for 93% of the total addressable market (TAM). Whether Wall Street believes it is another matter, but Musk can tell this story smoothly and calmly, showing his superb storytelling ability and excellent psychological quality.

PS: I'm reminded of the well - known bad joke that "Musk has a high IQ but a low EQ." I think the inventors and spreaders of this bad joke are destined to be nailed to the pillar of shame in history. However, this might be their only chance to be remembered in history, after all, a bad reputation is still a reputation.

Now, let's get serious. I'm currently paying for three foundation models simultaneously: GPT, Claude, and Grok. I also paid for Gemini for a while, but I've stopped now. There's not much to say about GPT. I've been an old user since 2023, with an uninterrupted payment history of 25 months. I'm already used to discussing all kinds of things with it. There's also not much to say about Claude. In the wave of Agentic Workflow, it would be abnormal not to use it (although I usually spend very little time using agents). That leaves Grok: its cost is significantly higher, with a basic membership costing $30 per month (the other two only charge $20), and its model capabilities are significantly weaker. If I want to cut one more paid model, it will definitely be Grok, not the other two.

Globally, especially in the United States, Grok is a foundation model in an awkward position:

• On the consumer side (C - end), GPT and Gemini have a significantly stronger user base. The former has a first - mover advantage, and the latter has the support of the Google ecosystem. Grok's only strength is its combination with X (Twitter), but how can the latter be compared with Google?
• On the business side (B - end), both Claude and GPT can outperform Grok by a large margin. Their capabilities are not on the same level. In terms of cost - effectiveness, it can't compete with several open - source models from China, including but not limited to DeepSeek.
• In terms of multi - modality, Grok obviously can't compare with Google's well - developed model series and doesn't even rank in the video field. X's data is mainly text, which may not be very helpful for training multi - modal models.

For a while, Grok took the open - source route and even once took over the torch of "the strongest open - source model in North America" from LLaMA. However, since Grok 2.5, the open - source route has been completely abandoned, while Google has returned to the open - source track through the Gemma series. Unless Musk has a change of heart, it's unlikely that Grok will become the flag - bearer of open - source.

So, where exactly are Grok's advantages? Why do some people still pay for it? There's an open secret: Fan fiction circles and writers (and their readers) who engage in borderline content love to use Grok because its content review criteria are very lenient, which is better than nothing. Claude is most afraid of users generating "unhealthy content" and is extremely strict, fearing that you'll cross the line in any way. GPT has tightened its criteria further since entering the 5.0 era. Gemini is much more lenient, but it doesn't compromise on principle. Only Grok is literally "utterly unscrupulous." As long as you ask, it will answer.

Additionally, Gemini, which is also relatively lenient, has a major problem: the hallucinations generated by context association are too serious, and the value of long - term conversations is relatively low. Those who use Gemini to analyze long texts know that it's best to feed the entire text to it at once, not paragraph by paragraph, otherwise, it will generate many strange cross - hallucinations. It has high value for brainstorming but no professional value. This further highlights the significance of Grok. At least it can be used in the field of long texts.

In terms of "anti - political correctness," Grok is also unique: each new version of GPT becomes more left - leaning, and Claude is inherently left - leaning. Compared with them, Grok can be regarded as the only right - wing leading large model. Especially for some topics involving the conservative - liberal binary opposition, the answers of other large models are almost unreadable. Grok at least tries to maintain nominal neutrality. As for whether this reflects Musk's own ideology, I have no idea.

I'm also a bit surprised by Grok's rich knowledge in the fields of anime and games. Whenever I play a game or watch an anime, I'll definitely discuss it with Grok first, but not for other topics. I think Grok seems to have formed a deep imprint of anime culture, not just a simple accumulation of corpora. As we all know, Musk is an anime lover and also likes to play games. I have no idea whether Grok's above - mentioned characteristics are related to this.

Besides that, there's a technical feature: Grok's search ability seems to be particularly powerful. It can directly read a large amount of web information, even information behind some "paywalls." Try asking GPT, Claude, Gemini, and Grok simultaneously about the latest report from a European or American paid media like the WSJ. Grok seems to be the only large model capable of breaking through the "paywall" and directly citing the original text. I believe that many users have "freeloaded" a large amount of information that would otherwise require payment in this way. Perhaps this is one of Grok's core competitiveness?

However, this doesn't mean that Grok has some "black technology" that others don't have. I think the most reasonable explanation is that Grok's target is not big enough, so not many people are watching these "gray - area" things. If the same thing happened to GPT or Claude, it would probably become global headlines tomorrow and be sued for billions of dollars. Assuming Grok develops hundreds of millions of paying users, the above - mentioned "abilities" will definitely be changed or at least greatly restricted.

Actually, Grok's other "advantages" can probably be explained in the same way. The content criteria of GPT and Gemini are becoming stricter because their targets are too big, and they need to be on high alert. (Note: Claude is an exception. Its content criteria have been strict from the beginning, purely because the management is abnormal, taking pleasure in torturing users, commonly known as "being arrogant with talent." This term was used to describe Yang Xiu in the famous Chinese classical novel "Romance of the Three Kingdoms.") That is to say, all of Grok's advantages are based on the fact that it's not big enough... Although based on the current situation, its hope of growing further is not high.

During the IPO roadshow, SpaceX clearly stated that it will "start deploying space computing power in 2028" and believes that "orbital AI computing power is technically ready, cheaper, and faster" (Readily available, cheaper, and faster). In the long term, SpaceX positions itself as a "space technology infrastructure provider," and this positioning doesn't necessarily require self - developed large models. As we all know, currently, the world's most advanced AI cloud service providers are Microsoft and Amazon, and both of their self - developed large models are relatively weak. Oracle, which has been rising in the past two years, has hardly wasted resources on large models.

In the near term, there are also obvious signs that xAI (which no longer exists and has completely become several departments of SpaceX) is shifting towards the computing power provider model: it has signed computing power leasing contracts worth $1.25 billion and $920 million per month with Anthropic and Google respectively. Musk once proposed a money - making formula: "Buy a lot of GPUs and then make money" - there's no need for the step of "through its own large model/application layer" in the middle. Anyway, the AI business models of Anthropic and Google have already been proven successful. Who cares whose money to make?

There's only one uncertain factor here: SpaceX has an option to acquire Cursor for $6 billion, and Cursor is the most popular AI programming tool. Coding & Agent is currently the AI track with the largest token consumption and the most certain business model. Musk obviously hopes to keep a ticket for this track, and it's the best ticket available at this stage because neither Claude Code nor GPT Codex is for sale.

But still, this doesn't require maintaining a powerful self - developed large model. Cursor is originally "model - neutral." Many people use it to run Claude, and some use it to run GPT. The model R & D strength of the Cursor team is barely there. They once developed a "self - developed model," which was later proven to be a modified version of Kimi. This also shows from one side that whether to have a first - party base model is not important for a tool like Cursor. We can even think that this is a symbol of SpaceX's AI business fully shifting towards "infrastructure": computing power is provided as an infrastructure to all users, and so are tools, as long as the price is right.

Anyway, I still hope that Grok can keep operating. It doesn't have to become one of the most powerful models in the world, but it must exist. I really enjoy discussing the plots and characters of anime works like "Zombie Land Saga," "Daredevil Party," and "Sound! Euphonium" with Grok. Its insights have benefited me a lot. It would be even better if the monthly price could be halved. I hope Musk will never consider shutting down Grok!

This article is from the WeChat official account "Internet Phantom Thieves" (ID: TMTphantom). The author is Pei Pei, the leader of the Phantom Thieves. It is published by 36Kr with authorization.