SpaceX (SPCX.US) accelerates AI infrastructure deployment! Colossus leads the global computing power rankings, Musk says space AI computing power will surpass terrestrial data centers by the end of 2028

The Zhitong Finance App learned that SpaceX (SPCX.US) CEO Elon Musk predicts that by the end of 2028, the artificial intelligence (AI) computing power deployed in orbit may exceed the additional terrestrial computing power. This ambitious vision shows SpaceX's plans to expand into the AI infrastructure sector, and its superintelligence (SI) business is competing with Microsoft (MSFT.US), Google (GOOGL.US), Amazon (AMZN.US), and OpenAI.

Musk posted on social networking platform X on October 30 that the monthly new orbital computing power may surpass the new terrestrial computing power around the end of 2028, and that this gap will widen significantly in 2029. “Compared to orbital computing power, terrestrial computing power will be insignificant,” he said.

SpaceX is betting big on space computing power! Rail data centers may usher in real scale in the 2030s

At the beginning of 2026, SpaceX revealed a highly disruptive space infrastructure construction plan and applied for permission from the US Federal Communications Commission (FCC) to launch 1 million data center satellites. SpaceX will deploy up to 1 million satellites within 500 to 2,000 km of low-Earth orbit. It plans to use it to build an orbital data center system with ultra-large-scale computing capabilities, directly targeting AI model inference, machine learning, edge computing, and related applications.

Building AI orbital data centers in space can help overcome the bottlenecks of terrestrial AI data centers, including tight power supply, high cooling costs, and large construction land. SpaceX believes that if Starship (Starship) can be used for large-scale low-cost launches, combined with Starlink (Starlink)'s accumulated satellite manufacturing, interplanetary laser links, and constellation operation capabilities, it is possible to further upgrade the “communication constellation” to a “computational constellation.”

At the end of August, Musk announced on the X platform that SpaceX's first Starmind AI satellites equipped with Nvidia (NVDA.US) chips will be launched for the first time in the fourth quarter of 2027 and will reach a “significant scale” in 2028. This schedule is ahead of the 2028 plan previously disclosed in the prospectus.

On the same day, Nvidia officially announced that SpaceXAI will deploy its next-generation Vera CPU to accelerate agent-based AI (Agentic AI) workloads and launch an optimized version of the Vera Rubin NVL72 rack-scale system into orbit as the computing power core of the first-generation Starmind satellite. This collaboration marks a critical turning point from concept to engineering implementation for the world's first space AI data center project.

Rail data centers will help support large-scale AI construction, but before they are actually practical, a series of major technological breakthroughs are needed. Although Musk plans to launch the first Starmind AI satellite by the end of 2027, other industry observers believe that actual large-scale deployment is more likely to take until the 2030s.

For SpaceX, there are a series of challenges that need to be overcome. The first is heat dissipation. Terrestrial data centers can rely on liquid cooling to manage heat, but in a space vacuum environment, cooling works quite differently. The second is large-scale data transmission. Transcelestial, a company focused on laser communications, is working to solve a key challenge — how to send massive amounts of data back to Earth. There's also how to power hyperscale deployments.

Additionally, the rail data center's cost ledger is being closely scrutinized by Wall Street. Business consulting agency Wood Mackenzie estimates that the cost of building a 1 gigawatt orbital data center is about 170 billion US dollars, which is more than three times that of a traditional terrestrial data center. Morgan Stanley analyst Adam Jonas predicts that SpaceX's AI-related capital expenditure will reach 53 billion US dollars in 2026, and an increase of 130 billion US dollars in 2027. Blue Origin, Amazon founder Jeff Bezos, and researcher Andrew McCalip pointed out that expensive AI chips and high launch costs constitute a hindrance to the current industry, and the current economic model is not reasonable.

Proponents, however, believe that space data centers can circumvent regulatory hurdles and community backlash from terrestrial data centers. At the same time, SpaceX is trying to build a vertically integrated supply chain to break down cost barriers and promote economic improvements in orbital computing. On the one hand, they rely on Starship heavy rockets to drastically reduce launch costs, and on the other hand, they jointly develop and manufacture their own AI chips with partners Tesla and Intel through a chip factory called Terafab. By controlling the launch and underlying computing power hardware, SpaceX is accelerating the commercialization process of its space computing.

Colossus leads the AI computing power ranking

When Musk made these remarks, a ranking of major AI computing power facilities compiled by Epoch AI showed that SpaceX's SI business's Colossus 2 facility in Memphis is currently at the top of the list. The facility is expected to have computing power equivalent to 1.11 million NVDA.US (NVDA.US) H100 GPUs, higher than the 1.02 million H100 equivalent GPUs owned by OpenAI's Stargate project in Abilene, Texas, and about 818,000 H100 equivalent GPUs owned by the project between Amazon and Anthropic in New Carlisle, Indiana.

H100 equivalent computing power is a standardized measurement of AI computing power. The performance of the Nvidia (NVDA) H100 graphics processor (GPU) is used as a benchmark to estimate the computing power that different processors can provide. The H100 is a chip widely used to train and run AI models.

For example, a data center with a nominal H100 equivalent computing power of 1 million H100 may not have actually installed 1 million H100 chips. It may also use a newer generation or a different type of processor, and the combined estimated performance of these processors is comparable to the above benchmarks. This metric helps compare data centers using different generations of hardware; however, actual performance will still vary depending on workload and system configuration.

Epoch AI estimates that as more computing clusters are put into operation, the computing power of Colossus 2 may reach about 1.82 million H100 equivalent GPUs by the beginning of 2027. The estimates are based on satellite imagery, cooling infrastructure, and information disclosed by the company, rather than a full physical inventory of installed processors.

Although AI computing power is rapidly expanding, there is still uncertainty about its economic benefits. In another discussion on the X platform, Musk mentioned restrictions on the use of Grok Bot and acknowledged that the cost of operating this service was too high compared to the revenue it generated, making it difficult to sustain. This statement highlights one of the core issues facing AI infrastructure investors: can the growing demand for computing power generate sufficient revenue, thus justifying the huge investments made to meet these needs.

Earlier, venture capitalist Shaun Maguire released a financial forecast that predicts SpaceX's revenue will soar from $51.5 billion in 2026 to $274 billion in 2027, and reach $731 billion in 2028. For investors, these predictions highlight the huge potential revenue opportunities in the AI infrastructure sector, and also reflect the capital investment and implementation risks required to build related infrastructure.

Shaun Maguire's model assumes that ground computing power will expand from 1 gigawatt at the beginning of 2026 to 21 gigawatts at the end of 2028, and each gigawatt can generate about 40 billion US dollars in revenue per year. The model predicts SpaceX's profit before interest, tax, depreciation and amortization (EBITDA) in 2028 will reach $536 billion. However, since capital expenditure for the year is expected to reach 369 billion US dollars, free cash flow will not be corrected until the middle of the year.

Shaun Maguire is a partner at Sequoia Capital, and the agency holds a large amount of shares in SpaceX (SPCX). He cautioned that there is considerable uncertainty in its predictions due to possible political hurdles and supply chain restrictions. Musk suggests, however, that these predictions may still be conservative. This means that even investors with a financial interest in the company may be underestimating SpaceX's growth potential.

Musk also raised concerns about the high cost of building AI infrastructure. Kyle Reidhead, an analyst and CEO of the investment communications agency Milk Road, previously released another forecast, believing that SpaceX may invest 470 billion US dollars in capital expenditure over five quarters and accumulate about 150 billion US dollars in debt. In response, Musk said that the market “seriously underestimated” the company's capital efficiency. Neither of these forecasts represents the company's published financial results or official performance guidelines.