SpaceX is reportedly racing to build the world’s largest AI cloud
SpaceX may be moving faster than any other company in the world when it comes to building AI data centers. After entering the artificial intelligence infrastructure market through its merger with xAI earlier this year, the Elon Musk-led company is now being described as a potential future leader in AI cloud computing.
According to research firm SemiAnalysis, SpaceX already has around 3.6GW of AI data center capacity under construction across multiple locations. The firm also claims it has identified more than 9GW of planned capacity as SpaceX pushes toward a revised goal of roughly 13GW by 2027.
If SpaceX reaches that target, it would represent a massive jump in a very short period of time. SemiAnalysis noted that the amount of capacity SpaceX could bring online in about a year and a half would be comparable to the total data center capacity Google built over more than two decades.
Elon Musk’s AI compute ambitions keep growing
SpaceX’s AI infrastructure plans appear to be expanding quickly. During the company’s first earnings call as a public company in August, Elon Musk said SpaceX expected to finish the year with more than 2GW of compute capacity. That was already a sharp increase from 1.4GW at the end of the previous quarter and just 400MW a year earlier.
At the time, Musk suggested that SpaceX could have somewhere closer to 10GW of cumulative compute online by the end of 2027, rather than 5GW. But by late October, the company’s ambitions had reportedly grown again. SpaceX was said to be planning an additional 10GW of compute capacity in 2027 alone, which could bring its total to at least 13GW, supported by as much as 20GW of power and cooling infrastructure.
From Grok training clusters to a major AI cloud business
SpaceX’s move into AI infrastructure began with its February merger with xAI, creating a combined company valued at about $1.25 trillion. The data centers in Memphis, originally designed to support Grok, quickly became valuable assets for outside customers as demand for AI compute continued to surge.
In May, Anthropic agreed to use the full compute capacity of Colossus 1. That deal gave the Claude developer access to more than 300MW of new capacity powered by over 220,000 Nvidia GPUs.
In June, shortly before SpaceX’s IPO, Google reportedly agreed to pay $920 million per month from October 2026 through June 2029 for capacity based on around 110,000 Nvidia GPUs. The agreement was described as a way to secure temporary capacity to meet rising demand for Google’s Gemini Enterprise agent platform.
SpaceX CFO Bret Johnsen said leasing compute capacity to customers such as Anthropic, Google, and Reflection AI helped generate about $1.6 billion in revenue.
SpaceX is cutting data center build times dramatically
One of the biggest reasons SpaceX is attracting attention in the AI infrastructure race is speed. SemiAnalysis said many data center developers are struggling to complete projects in eight to twelve months, while SpaceX is reportedly cutting that timeline roughly in half.
The firm attributes part of that advantage to the engineering experience across Musk’s companies. It also believes Tesla may be producing modular mechanical, electrical, and plumbing equipment for SpaceX’s AI data center projects, helping accelerate construction.
Musk has argued that SpaceX’s experience with rockets gives it a major advantage in building complex infrastructure quickly. He has compared applying rocket engineering to data centers as an overwhelming mismatch against traditional builders.
SpaceX’s Texas expansion could include more data centers
SpaceX’s Starbase facility in Texas, known for rocket development and launches, is also expected to include two data centers, according to SemiAnalysis. The research firm also believes there is a strong possibility that Terafab, another major Musk-linked project, could include a data center as well, potentially with capacity leased to outside customers.
Texas Governor Greg Abbott previously said SpaceX would invest $16.8 billion in the first phase of Terafab, a massive 100-million-square-foot campus in Grimes County. The site is expected to produce chips for SpaceX and Tesla.
For now, Musk has said SpaceX will rely exclusively on Nvidia hardware for its AI compute buildout.
Power supply remains the biggest challenge
While SpaceX may be building AI data centers at remarkable speed, power availability remains a major obstacle. SemiAnalysis said SpaceX’s sites are going up so quickly that the company sometimes identifies the power supply while construction is already underway.
The firm is tracking more than 8GW of power under Musk’s management, which is still below the more than 9GW of data center capacity it believes is coming online.
At Colossus 2, turbines at the Southaven power plant can legally operate at 2.7GW of nameplate capacity today, but that figure is expected to fall to 1.24GW by July 2027. That could leave a potential gap of up to 300MW, which SemiAnalysis suggests may be addressed with a large battery storage deployment.
Can SpaceX really become the world’s biggest AI cloud?
SpaceX has a long history of setting aggressive goals, and Musk’s companies are known for ambitious timelines that are not always met. Still, the scale and speed of SpaceX’s AI data center expansion are difficult to ignore.
If the company can secure enough power, cooling, chips, and customers, SpaceX could become one of the most important players in global AI infrastructure within just a few years. What began as compute for Grok may now be turning into a much larger business: a high-speed AI cloud platform built around Nvidia GPUs, massive energy needs, and Musk’s willingness to rethink how data centers are built.
