The mystery of SpaceX's "leasing out computing power" has been explained: "technical issues" prevent meeting model training needs, so all computing power of Colossus 1 is being leased out.

The mystery of SpaceX's "leasing out computing power" has been explained: "technical issues" prevent meeting model training needs, so all computing power of Colossus 1 is being leased out.

```

SpaceX is renting out all computing power of its Memphis data center Colossus 1, and the real reason has finally surfaced—this is not a proactive strategic shift, but a pragmatic monetization forced by technical limitations.

According to the latest report from Bloomberg, insiders revealed that SpaceX encountered serious latency issues when attempting to network Colossus 1 with two other data center campuses to collaboratively train AI models. Coupled with aging network infrastructure, this ultimately rendered the facility unable to meet the technological requirements for training cutting-edge large models. Against this backdrop, SpaceX decided to lease out Colossus 1 as a whole, converting idle computing power into a stable source of revenue.

WallstreetCN previously mentioned that SpaceX has successively signed massive computing power leasing agreements with Anthropic and Google, with annualized income reaching about $26 billion, which has become the core narrative supporting its $2 trillion valuation. However, critics pointed out: the transaction pricing exceeds the market rate and is suspected of "circular financing," claiming this is merely xAI's "mess monetization" after experiencing GPU architecture chaos and setbacks in model training.

Undoubtedly, this disclosure exerts subtle pressure on SpaceX's narrative. The technical predicaments behind the leasing of Colossus 1 prompt the outside world to rethink the company's capability to rapidly expand its data center infrastructure.

Latency and hardware mixing, Colossus 1 loses the chance for large model training

According to Bloomberg quoting insiders, SpaceX initially planned to use a super-large cluster composed of three data center campuses to train its most advanced AI models. However, there were significant network latency issues between Colossus 1 and the other two sites, which are over 10 miles apart. Aging network infrastructure further exacerbated this bottleneck.

Training bigger and stronger AI models requires ultra-fast inter-node connections. If bandwidth between facilities is insufficient or there is latency, the entire cluster's training efficiency is dragged down to the level of the slowest node.

The issue of heterogeneous hardware is also not to be ignored. Insiders revealed that Colossus 1 internally deployed a mix of multiple NVIDIA chip generations, including Hopper and Blackwell systems, as well as some older generation accelerators. In contrast, Colossus 2 and Colossus 3 were deployed more uniformly around NVIDIA Blackwell chips. In a distributed training cluster, older chips become bottlenecks, forcing faster accelerators to wait, causing overall performance to match the slowest hardware.

Faced with these technical limitations, SpaceX eventually determined it would be better to lease out Colossus 1's capacity to external customers, rather than continue pouring resources into addressing these issues, while reserving newer facilities for its own AI development.

$26 billion annualized contracts, monetizing computing power to support IPO narrative

According to Reuters, SpaceX has signed massive computing power leasing agreements with Anthropic and Google—Anthropic pays $1.25 billion per month, and Google pays $920 million per month starting this October. The combined annualized revenue scale of these two contracts is about $26 billion, with the total contract value exceeding $70 billion.

This series of deals provides a strong revenue narrative for SpaceX's IPO targeted at raising as much as $75 billion. In its roadshow, SpaceX highlighted data center construction as a key selling point for investors, emphasizing that its first Colossus facility was completed in just 122 days, surpassing both its own expectations and industry averages.

SpaceX CFO Bret Johnsen recently stated that the company has not abandoned its internal AI services, including Grok. Elon Musk himself also stated that SpaceX reserves the right to terminate the Anthropic computing power agreement ahead of schedule, saying, "If computing power becomes extremely tight, I've said we may have to take it back at some point."

The tension between technical dilemmas and the IPO narrative

The Colossus 1 leasing incident reveals not only the technical limitations of a data center, but also highlights the deeper challenges SpaceX faces in rapidly expanding its AI infrastructure.

Earlier this year, SpaceX completed the acquisition of Elon Musk's xAI and positioned its data center business as a strategic pivot toward becoming an AI infrastructure provider. However, because Colossus 1 could not be used for cutting-edge model training due to hardware mixing and network latency, doubts have arisen over the credibility of this strategic narrative.

Market observers noted that renting out computing power to competitors instead of using it to train their own models suggests that internal AI development has encountered setbacks. Meanwhile, some analysts questioned whether the transaction pricing exceeds the market rate and whether the deal structure displays "circular financing" characteristics.

SpaceX’s official name is Space Exploration Technologies Corp. Its representative did not respond to Bloomberg's request for comment.

Risk warning and disclaimerThe market involves risks, and investments should be made cautiously. This article does not constitute personal investment advice, nor does it consider individual users’ special investment goals, financial situations, or needs. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investing accordingly is at your own risk. ```