SpaceX (SPCX.US) CEO Elon Musk predicts that by the end of 2028, AI compute deployed in orbit could exceed new terrestrial compute capacity.
This ambitious vision showcases SpaceX's plans to expand into the AI infrastructure sector, with its Superintelligence (SI) business competing against Microsoft (MSFT.US), Google (GOOGL.US), Amazon (AMZN.US), and OpenAI.
Musk posted on social platform X on October 30 that monthly新增 orbital compute could surpass new terrestrial compute around the end of 2028, and the gap will widen significantly in 2029.
He stated: "Compared to orbital compute, terrestrial compute will be negligible."
SpaceX's Bold Bet on Space Compute! Orbital Data Centers May Not Achieve True Scale Until the 2030s
In early 2026, SpaceX disclosed a highly disruptive space infrastructure plan, filing with the U.S. Federal Communications Commission (FCC) for a license to launch 1 million data center satellites.
SpaceX plans to deploy up to 1 million satellites in low Earth orbit between 500 and 2,000 kilometers, aiming to build an orbital data center system with hyperscale computing capabilities, directly targeting AI model inference, machine learning, edge computing, and related applications.
Building AI orbital data centers in space helps overcome bottlenecks facing terrestrial AI data centers, including tight power supply, high cooling costs, and large land requirements.
SpaceX believes that if it can achieve large-scale, low-cost launches with Starship, combined with Starlink's accumulated satellite manufacturing, inter-satellite laser links, and constellation operations capabilities, it could upgrade its "communications constellation" into a "computing constellation."
In late August, Musk announced on X that SpaceX's first batch of Starmind AI satellites carrying NVIDIA (NVDA.US) chips will launch in the fourth quarter of 2027, reaching "significant scale" by 2028.
This timeline is earlier than the 2028 plan previously disclosed in its prospectus.
On the same day, NVIDIA officially announced that SpaceX AI will deploy its next-generation Vera CPU to accelerate Agentic AI workloads, and will send an optimized Vera Rubin NVL72 rack-scale system into orbit as the computing core of the first-generation Starmind satellites.
This partnership marks a key turning point for the world's first space AI data center project, moving from concept to engineering implementation.
Orbital data centers will help support large-scale AI buildout, but before they become truly practical, a series of major technological breakthroughs are still needed.
Although Musk plans to launch the first Starmind AI satellites by the end of 2027, other industry observers believe that true large-scale deployment is more likely to wait until the 2030s.
For SpaceX, a series of difficulties must be overcome.
First is heat dissipation. Terrestrial data centers can rely on liquid cooling to manage heat, but in the vacuum of space, heat dissipation works completely differently.
Second is large-scale data transmission. Transcelestial, a company focused on laser communications, is working to solve a key challenge—how to transmit massive amounts of data back to Earth.
There is also the question of how to provide power for hyperscale deployments.
In addition, the cost equation for orbital data centers is under close scrutiny from Wall Street.
Business consulting firm Wood Mackenzie estimates that building a 1-gigawatt orbital data center costs about $170 billion, more than three times that of a traditional terrestrial data center.
Morgan Stanley analyst Adam Jonas expects SpaceX's AI-related capital expenditure to reach $53 billion in 2026 and increase by another $130 billion in 2027.
Blue Origin and Amazon founder Jeff Bezos, along with researcher Andrew McCalip and others, have pointed out that expensive AI chips and high launch costs constitute current industry obstacles, and the economic model at this stage is not yet reasonable.
However, supporters argue that space data centers can circumvent regulatory obstacles and community backlash faced by terrestrial data centers.
At the same time, SpaceX is trying to build a vertically integrated supply chain to break through cost barriers and improve the economics of orbital computing.
On one hand, it relies on the Starship heavy-lift rocket to significantly lower launch costs; on the other hand, through a chip factory called Terafab, it is jointly developing and manufacturing its own AI chips with partners Tesla and Intel.
By controlling launch and underlying computing hardware, SpaceX is accelerating the commercialization of its space computing business.
Colossus Leads AI Compute Rankings
As Musk made these remarks, a ranking of major AI compute facilities compiled by Epoch AI shows that SpaceX's SI business's Colossus 2 facility in Memphis currently ranks first.
The facility is expected to have compute equivalent to 1.11 million NVIDIA (NVDA.US) H100 GPUs, higher than the 1.02 million H100-equivalent GPUs of OpenAI's Stargate project in Abilene, Texas, and the approximately 818,000 H100-equivalent GPUs of the project jointly built by Amazon and Anthropic in New Carlisle, Indiana.
H100-equivalent compute is a standardized AI compute measurement metric that uses the performance of NVIDIA's (NVDA) H100 graphics processing unit (GPU) as a benchmark to estimate the compute that different processors can provide.
The H100 is a chip widely used to train and run AI models.
For example, a data center nominally having 1 million H100-equivalent compute does not necessarily actually have 1 million H100 chips installed.
It may also use a newer generation or different type of processor, whose combined estimated performance is comparable to the above benchmark.
This metric helps compare data centers using different generations of hardware, although actual performance will still vary depending on workloads and system configurations.
Epoch AI estimates that as more computing clusters come online, Colossus 2's compute could reach about 1.82 million H100-equivalent GPUs by early 2027.
Its estimates are based on satellite imagery, cooling infrastructure, and company disclosures, rather than a comprehensive on-site inventory of installed processors.
Although AI compute is expanding rapidly, its economic benefits remain uncertain.
In another discussion on X, Musk acknowledged while discussing usage limits for Grok Bot that the cost of operating the service is too high relative to the revenue it generates to be sustainable.
This statement highlights a core issue facing AI infrastructure investors: whether growing compute demand can generate enough revenue to justify the massive investments made to meet that demand.
Previously, venture capitalist Shaun Maguire released a financial forecast predicting that SpaceX's revenue will surge from $51.5 billion in 2026 to $274 billion in 2027, and reach $731 billion in 2028.
For investors, these projections highlight the enormous potential revenue opportunity in the AI infrastructure sector, while also reflecting the capital investment and execution risks required to build related infrastructure.
Shaun Maguire's model assumes that terrestrial compute will expand from 1 gigawatt in early 2026 to 21 gigawatts by the end of 2028, with each gigawatt generating about $40 billion in annual revenue.
The model projects that SpaceX's 2028 EBITDA will reach $536 billion.
However, because annual capital expenditure that year is expected to be as high as $369 billion, free cash flow will not turn positive until mid-year.
Shaun Maguire is a partner at Sequoia Capital, which holds a large stake in SpaceX (SPCX).
He cautioned that his forecasts carry considerable uncertainty due to potential political obstacles and supply chain constraints.
However, Musk hinted that these projections 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 questioned外界 concerns about the high cost of building AI infrastructure.
Kyle Reidhead, an analyst and CEO at investment newsletter firm Milk Road, previously released another forecast suggesting SpaceX could spend $470 billion in capital expenditure over five quarters and accumulate about $150 billion in debt.
In response, Musk said the market "severely underestimates" the company's capital efficiency.
Neither of the above forecasts represents the company's published financial results or official guidance.
Comments