The cost is so high it's unsustainable—so why is Musk still doubling down on Grok Bot?

Original | Odaily@OdailyChina)

Author | Wenser (@wenser 2010)

If you follow Musk's X platform account, you'll notice that his mentions of "Grok Bot" have recently reached a fever pitch, rivaling even the frequency of his posts ahead of Tesla and SpaceX product launches. Such high-frequency promotion has naturally drawn more and more attention and usage to Grok Bot, and with that come controversy and skepticism.

On November 1, someone called out to Musk, "Grok Bot's quota has been depleting especially fast lately—when are you going to do a quota reset?" Musk responded that "the cost of providing Grok Bot's services is unsustainably high relative to revenue," and stated that "even though the era of AI abundance will eventually arrive, unlimited usage is simply not feasible at this stage"; he also noted that whether users find the high-quota plan's price increase too expensive, and whether to lower quotas with advance notice, are all matters that require trade-offs.

From publicly talking down the new cloud vendors a while back to his recent continued bets on and endorsement of Grok Bot, what this reflects is not just Musk's optimism and affirmation of Grok Bot, but also his larger ambition to "seize the AI high ground." Today, let's look from a different angle at whether Musk's "AI layout" can deliver results.

Behind Grok Bot's Explosive Popularity: The Rise of Personal Agents, the Cursor Team Going All Out

At present, we can roughly divide the AI industry into 3 layers:

• First, infrastructure such as chips, computing power, and energy/electricity;
• Second, large models such as OpenAI's GPT Astra and Anthropic's Claude Opus 5.5;
• Third, Agent applications such as Meta's Muse and Grok's Grok Bot.

Musk has laid out plans across all 3 of these areas, with his focus constantly adjusting and developing in coordination. Specifically, his layout can be divided into the following 3 parts:

First, the chip and computing power layer, which leans toward infrastructure construction. In this regard, Musk's recent mention of escalating competition in the cloud vendor space—where new players may face accelerated elimination—is a clear strategic play. For more details, see "Talking Down New Cloud Vendors: Musk's Top-Level Strategic Play." As for his layout: on one hand, there are SpaceX's past reserves of power, energy, and computing chips, and the previously reached cooperation with Anthropic is also an important part of its strategic transformation; on the other hand, the future development roadmap has already been mapped out: Musk previously boldly claimed that space-based computing power is expected to surpass ground-based computing power by the end of 2028; in addition, Terafab, a chip manufacturing joint venture between SpaceX and Tesla, plans to achieve first-batch chip mass production in Q3 2028. Controlling infrastructure is equivalent to locking in key resources in advance, ensuring at least an invincible position.

Second, the AI model layer, where AI giants currently gather. In the past, Musk had high hopes for Grok, but over time, the Grok model has consistently struggled to catch up with the latest models from OpenAI and Anthropic in industry competition. Ultimately, although the Grok model team was retained, in May of this year Musk still made a "alliance-building" decision—teaming up with Anthropic to counter OpenAI. For more details, see "Musk and Anthropic Are Going to Space to Find Power."

Finally, there is the increasingly mature Agent application layer, following the "OpenClaw lobster craze" earlier this year. Meta's Muse and the previously controversy-hit Manus were pioneers, providing a certain degree of user education and scenario breakthroughs for Agent applications. After the Cursor team successfully merged into SpaceXAI, Grok Bot became the team's passion project. It is precisely under the push of the Cursor team, which pursues extreme user experience, that Grok Bot is now advancing rapidly through fast iterative updates, while leveraging the data advantages of the X platform to further fuel Grok Bot's scenario expansion and user growth momentum.

In summary, SpaceX's successful listing this year, the lavish acquisition of Cursor for over $60 billion in stock, and a series of prior accumulations have opened new possibilities for Musk's "infrastructure + model + application three-step AI strategy." With the help of the hit application Grok Bot, SpaceXAI—Musk's "AI mega-group organization"—has for the first time achieved the possibility of connecting the AI industry's upstream and downstream through a curved path to seize the high ground in the sector's development.

After all, no matter how powerful the infrastructure resources or how formidable the AI large models, an application is always needed as a channel and entry point to reach users and open markets.To some extent, this is the "new solution for AI industry layout" that Musk found after ending the litigation dispute with OpenAI.

The Potential Hidden Concerns Behind Musk's "Three-Line Layout": Infrastructure Stability, the Singularity Threat of Large Models, and White-Hot Application Competition

Of course, at a time when Anthropic is about to IPO and AI sector stocks are undergoing a sharp correction, this does not mean Musk and SpaceXAI can already rest easy. Because, in terms of the industry's development stage, we haven't even reached the "second half of AI industry development"—not even the mid-stage. Therefore, if Musk and SpaceXAI want to occupy more advantageous positions and seize more initiative, they still need to face the following risks:

First, SpaceX's debt pressure and infrastructure stability assurance. Although Musk is full of confidence in space-based computing power and chip layout, given his past "credibility score," it is still hard to fully believe his "2028 pie in the sky"; his earlier mention that "SpaceX has completed $40 billion in GPU-related debt financing" encountered negotiation obstacles is also a prime example.

Second, the "singularity moment" threat from the chaotic battle of AI large models. If top-tier AI large models subsequently usher in a strong-update "singularity moment," with the flagship models of major leading AI vendors achieving historic capability breakthroughs, then after Grok's model capabilities fall out of the first tier, the market and users could withdraw at any time and flock to the newest and strongest AI large models.

Third, the more intense industry competition at the application level. The OpenClaw "lobster craze" at the beginning of the year burned hot, but it has now been integrated into the underlying rules of various applications and models; Work Agents and Personal Agents are merely different forms and applicable scenarios of Agent applications today, and more comprehensive Personal Assistant roles and applications will inevitably emerge. Not only are AI giants working on this, but internet tech giants are also competing hard, and individual independent developers and startups are targeting this market as well. Although the underlying models called upon differ only marginally, the differences in user experience and penetration at the application level are real. Grok Bot's current popularity says little in terms of user numbers and ROI input-output ratio.

Conclusion: After Work Agents and Personal Agents, the Next Step for AI Agents Is Life Agents

Finally, as a heavy user of Grok Bot, let me briefly share my current experience.

With AI large models having experienced nearly 4 years of rapid development, today's AI Agents have officially crossed the threshold from "intern" to "full-time employee." Junior employee tasks can already be confidently handed over to Grok Bot and Muse, as well as domestic applications like WorkBuddy and Qwen—as long as security management and disaster recovery backups are in place. So the next "battleground" for AI Agents is offline and online consumption scenarios—clothing, food, housing, transportation, dining, entertainment, travel, shopping, and so on.

This means not only a more complex and variable payment environment and payment amounts ranging from small to large, but also that the user market will impose higher demands on AI Agent applications' convenience, integration, high-frequency scheduling, and other aspects of performance.

Yesterday, the livestream of well-known e-commerce host Li Jiaqi also officially opened the buying and selling of "AI Token quota packages," which means "AI Tokens" are entering ordinary households just like mobile payment tools such as Alipay and WeChat Pay. What awaits Grok Bot and Musk is not only the broad proposition of infrastructure resource construction, but also more refined issues such as improving per-unit Token productivity and computing power conversion efficiency.

The high ground of the AI industry concerns not only the underlying large models—the applications that truly interact with frontline users also deserve great attention.