Anthropic may pay SpaceX US$84.5 billion for AI computing power, Musk's data center business receives huge orders

📅 2026-09-30

Abstract:

As artificial intelligence company Anthropic advances its initial public offering (IPO) plan, a newly disclosed document reveals a surprisingly large cooperation agreement with SpaceX. According to relevant documents reviewed by Reuters, Anthropic has signed a computing power purchase agreement worth up to US$84.5 billion and will pay SpaceX huge fees in the future to obtain the AI ​​computing power provided by its data center.

This number is much higher than the scale of cooperation previously known to the outside world, and it also allows the market to further see SpaceX’s ambition to expand in the field of artificial intelligence infrastructure. At the same time, Musk is accelerating the construction of large-scale AI data centers and plans to deploy more than 1 million GPUs by the end of this year to support the growing demand for artificial intelligence training.

In May of this year, SpaceX disclosed its cooperation with Anthropic for the first time when it submitted IPO-related documents. According to public information at the time, the agreement between the two parties covers the computing resources of the Colossus 1 and Colossus 2 data centers and is intended to help SpaceX commercialize the idle computing power in its infrastructure.

SpaceX stated at the time that Anthropic agreed to pay a monthly computing power fee of US$1.25 billion from May 2026 to May 2029, but lower charging standards would be adopted in May and June 2026 due to the production ramp-up stage. Based on this disclosure, the total value of the relevant agreements is approximately US$42.5 billion.

However, as the contents of Anthropic’s IPO documents were exposed, the outside world discovered that the actual scale of cooperation between the two parties was far greater than previously thought. Documents show that Anthropic has signed an agreement with SpaceX worth up to $84.5 billion to obtain data center computing resources based on Nvidia GPUs. The amount is about double the size previously disclosed.

HTZQP6VbMAAmIs5.png

It is worth noting that although the amounts of these agreements are large, most of the terms allow Anthropic to terminate the cooperation with 90 days' notice. This means that despite the long-term cooperation arrangement between the two parties, Anthropic retains some flexibility to adjust its infrastructure strategy based on future market changes.

The document also disclosed that Anthropic is expected to invest at least US$518 billion in building artificial intelligence infrastructure in the next ten years. In addition to SpaceX, its cooperation partners also include large cloud computing service providers such as Amazon, Alphabet's Google and Microsoft. As the demand for computing power from advanced AI models continues to rise, obtaining stable and sufficient GPU resources has become one of the key factors for industry competition.

For SpaceX, this potential revenue coincides with the rapid expansion of its AI infrastructure business. Musk is actively promoting the construction of the Colossus 2 data center and plans to have more than 1 million GPUs online before the end of the year. The project is called a "computing super factory" by the company, and most of the equipment will use Nvidia's latest generation Blackwell architecture GPUs.

After completing the integration of xAI earlier this year, SpaceX is gradually integrating and developing its rocket, satellite Internet and artificial intelligence infrastructure businesses. In addition to meeting the needs of its own AI projects, the company has also begun leasing large-scale computing resources to external customers, and Anthropic is one of the most important customers.

As the competition for generative artificial intelligence enters a new stage, the competition among major AI companies for high-performance computing resources continues to heat up. The huge computing power procurement plan disclosed by Anthropic not only reflects that the cost of training the next generation of artificial intelligence models has reached unprecedented levels, but also shows that data centers and GPU resources are becoming one of the most important strategic assets in the global technology industry.

Related tags

Related articles

Comments

0/500
Captcha (click to refresh)
No comments yet