Summary of Key Points
SpaceX has recently signed two high-value contracts for computing power leasing: Anthropic is paying $1.25 billion per month, and Google is paying $920 million per month, resulting in a monthly income of $2.17 billion, which amounts to $26 billion annually. The total value of these contracts exceeds $70 billion. This has significantly boosted SpaceX's upcoming IPO (Initial Public Offering) of $75 billion. However, on the other hand, its AI laboratory, xAI, is facing numerous challenges: a significant portion of its core talent has left, research and development efforts have stalled, and it has been accused of secretly using others' technology to train its models. This raises questions about SpaceX's true intentions—does it really want to develop its own AI capabilities, or does it simply want to make money by leasing computing power?
1. Two “Big Deals” That Boost the IPO
SpaceX’s latest contracts are truly lucrative:
- The Anthropic deal: Signed in May, it costs $1.25 billion per month and covers most of the computing power at the Colossus 1 data center in Memphis, Tennessee, for a minimum of six months. This data center is equipped with 220,000 NVIDIA GPUs, providing Claude, Anthropic’s chatbot, with 300 megawatts of computing power (equivalent to the electricity consumption of 300,000 households).
- The Google deal: Signed in June, it runs from October this year until June 2029 and costs $920 million per month for 110,000 NVIDIA GPUs and related resources. There is a condition, however: if SpaceX does not obtain the required number of GPUs by September 30, Google may terminate the contract or reduce the payment.
Together, these contracts generate a steady monthly income of $2.17 billion for SpaceX, amounting to $26 billion annually. If both contracts are fully executed, the total value will exceed $70 billion. This is crucial for SpaceX, as it needs substantial financial support to meet its $75 billion fundraising goal, and the revenue from these leasing agreements provides concrete evidence of its capabilities.
2. xAI’s Troubled Situation
In contrast to SpaceX’s successful leasing efforts, xAI (Musk’s AI laboratory) is in a mess:
- Severe Talent Loss: Many of the eight co-founders have left, and the engineering team has been downsized. The pre-training team, which once had over 20 people, now has fewer than five members. The person in charge of the Grok Code project has changed four times in just a few months, with the most recent departure occurring in May.
- R&D Challenges: An employee accidentally deleted the core training data for an AI programming product while migrating it, resulting in the loss of two to three weeks’ worth of work. Additionally, xAI encountered issues when trying to train the Grok model at the Colossus 1 data center due to the mixed installation of different GPU models (H100, H200, etc.), forcing them to move the training to the Colossus 2 data center.
- Financial Strains: xAI is a loss-making department that contributed significantly to SpaceX’s overall losses in the first quarter of this year. Without the revenue from Anthropic’s leases, the situation could have been even worse.
3. xAI’s Secret “Plagiarism” Allegations
What’s more embarrassing is the revelation that xAI has been using Anthropic’s technology:
- According to insiders, xAI launched a project that directly used responses from Anthropics’ Claude chatbot to train its own models.
- In January this year, Anthropic revoked xAI’s access rights, but the xAI engineers continued to use the data through personal accounts until they were blocked by Anthropic. xAI later attempted to access the data via a third-party platform, Blackbox AI, with the last usage in mid-May.
- Musk has previously acknowledged that xAI used some OpenAI models for training Grok, but not mentioned Anthropic’s involvement. This practice raises doubts about xAI’s ability to develop its own AI independently—how can it compete with companies like OpenAI and Anthropic if it relies on others’ technology?
4. Is Leasing Computing Power a “Casual Fortune”? xAI’s Mistakes Benefit SpaceX
SpaceX’s computing power leasing business only came into being because of xAI’s mistakes:
xAI intended to train the Grok model at Colossus 1, but due to the mixed installation of GPU models, the computing power was not effectively utilized and became idle. Seeing this opportunity, SpaceX decided to lease the resources to generate revenue, leading to these large contracts with Anthropic and Google.
Musk has also mentioned that these leases are temporary and that SpaceX may eventually retrieve the computing power. However, if it decides to withdraw the leased resources in the future, where will its income come from? This raises questions about whether SpaceX intends to rely on leasing for long-term revenue or truly develop its own AI capabilities.
5. The Glorious IPO, but Underlying Doubts
SpaceX’s IPO target of $75 billion is touted as one of the largest in U.S. history. However, there are concerns:
- Although leasing computing power is profitable, it represents the sale of resources rather than the development of AI products. If xAI fails to produce viable AI products, SpaceX’s “AI growth story” could prove unsustainable, as income from leasing is temporary, and independent AI capabilities are essential for long-term competitiveness.
- Musk has mentioned that xAI is working on the next generation of Grok models, but no timeline has been provided. If there is no progress after the IPO, will investors be disappointed?
In short, SpaceX’s current success relies on computing power leasing, but its ability to become a true AI competitor depends on xAI’s ability to resolve its internal issues. For outsiders, this news might suggest that Musk’s company has found another source of revenue, but its AI ambitions have encountered obstacles. Whether the IPO is worth investing in depends on whether SpaceX can overcome these challenges and develop its own AI capabilities.
Conclusion: While SpaceX’s computing power leasing business has enhanced its IPO prospects, xAI’s problems highlight its shortcomings in independent AI research and development. With substantial monthly income from leasing, but struggling R&D, it remains uncertain where SpaceX’s AI strategy is heading. For the general public, this news might indicate that Musk’s company has found another way to make money, but its AI aspirations have encountered difficulties. Whether the IPO is a worthwhile investment depends on whether SpaceX can overcome these obstacles and build a solid foundation in AI technology.