SpaceX is transforming its massive computing infrastructure into a high-speed revenue generator, yet the underlying contradictions in this business logic are drawing increasing market scrutiny.
According to reports from June 6th, SpaceX has signed substantial compute leasing agreements with Anthropic and Alphabet (GOOGL) – Anthropic will pay $1.25 billion monthly, while Alphabet will pay $920 million monthly starting this October. These contracts represent an annualized revenue stream of approximately $26 billion, with a combined total contract value exceeding $70 billion.
These deals provide powerful revenue narrative support for SpaceX's push towards what could be the largest IPO in U.S. stock market history, with a targeted fundraising size of up to $75 billion.
However, numerous market observers and analysts on social media are raising a more fundamental question: Why is SpaceX leasing its computing power to competitors instead of using it to train models for its own AI lab, xAI? The answer to this question may point to a costly design flaw within xAI, sparking a chain of doubts about the entire transaction's logic and the IPO narrative itself.
Concerns Over Compute Pricing and "Circular Finance"
Sharp questions have emerged on social media regarding the pricing rationale behind these two deals.
Former assistant research professor Roger posted a detailed mathematical breakdown of the Alphabet deal on platform X. He noted that CoreWeave's ($CRWV) pricing for GB200 compute is approximately $10.50 per hour, or $5.25 per hour per Blackwell GPU chip. Based on this, the annual cost for 110,000 chips would be around $5.06 billion.
In other words, Alphabet could obtain equivalent compute power from CoreWeave at a lower cost. Even if it paid a 100% premium to CoreWeave, it could still save about $1 billion—yet it chose to lease compute from SpaceX at a higher price.
Roger concluded that the prices Alphabet and Anthropic are paying SpaceX exceed the cost of building their own data centers. He characterized such transactions as a "circular finance" game where mega-cap tech companies use pension fund capital to inflate each other's valuations and create illusory revenue growth.
AI researcher Gary Marcus approached the issue from a different angle, arguing that focusing on how much SpaceX is charging is "asking the wrong question." He suggested the real question is why SpaceX made these deals. In his view, the answer is that xAI has realized it cannot win the cutting-edge model race, necessitating such transactions to maintain strong financial performance ahead of the IPO.
Internal Contradictions in the IPO Narrative and Orbital Data Center Vision
SpaceX's IPO filing lists "orbital data centers" as a core growth narrative, with plans for initial deployment around 2028. However, X user Chief Agenteer directly highlighted the logical gap between this narrative and reality.
He wrote, "SpaceX is leasing out idle compute, which in itself shows that orbital data centers are completely unnecessary and will never be economically viable. If they can't even fully utilize the compute they've built on the ground, why build more in space?"
This critique links two previously separate pieces of information: on one hand, SpaceX is forced to lease out ground-based compute due to xAI's training setbacks; on the other, its IPO filing simultaneously paints an ambitious picture of expanding compute capacity in space. The tension between these two points is causing some investors to question the overall strategic coherence of SpaceX's AI business.
xAI Training Setback: Colossus 1 Architecture Chaos as Catalyst
According to a report by tech media outlet WccfTech, SpaceX's xAI lab encountered severe technical obstacles at its Colossus 1 data center in Memphis, Tennessee. The data center deployed a mix of Nvidia H100, H200, and GB200 GPUs with different architectures. This heterogeneous design prevented xAI from effectively training its Grok model there, ultimately forcing the training workload to be migrated to the Colossus 2 data center.
This technical difficulty directly explains SpaceX's motivation for leasing out its compute. WccfTech's analysis suggests SpaceX is essentially trying to monetize xAI's poor design decisions—by signing cloud service agreements with external clients like Alphabet and Anthropic to turn the previously idle or inefficiently used Colossus 1 compute into significant rental income.
This background further fuels external doubts about xAI's independent R&D capabilities. SpaceX's AI lab is already facing dual pressures of core talent attrition and R&D setbacks, and the Colossus 1 architecture issues have compounded the problems.
SpaceX currently plans to complete its public listing next week. While the highly visible revenue from its compute leasing business provides strong support for the IPO, the controversies surrounding the transaction's pricing logic, xAI's R&D capabilities, and the commercial viability of orbital data centers may become variables that investors need to weigh carefully during the pricing process.
Comments