The Zhitong Finance App learned that the latest research report released by Deutsche Bank shows that the “AI+ Space Exploration” supergiant SpaceX (SPCX.US), founded and led by Musk, is rapidly shifting the focus of performance growth from the Starlink satellite internet platform to terrestrial AI computing power infrastructure. Deutsche Bank expects AI computing power leasing to contribute more than half of the tech giant's revenue in the fourth quarter; senior analyst Edison Yu from Deutsche Bank raised its quarterly revenue forecast to US$24.5 billion, which is about 8.4% higher than Wall Street's agreed estimate of US$22.6 billion, and the gross margin is expected to be close to 60%.
What supports Deutsche Bank's judgment is the delivery capacity and high rent of Colossus 2's ready-made super AI computing power cluster. In Musk's view, the space orbital AI data center is SpaceX's future core business, but in the opinion of Deutsche Bank analysts, it is an uncertain increase in the longer term — Deutsche Bank does not expect it to reach a real scale until 2029-2030.
Furthermore, Musk himself proposed at Barron's investment conference on November 6 that SpaceX's annualized revenue operating rate by December 2027 may exceed 500 billion US dollars, and made it clear that this is his latest personal judgment, not a specific growth promise; the 193.2 billion US dollar revenue data predicted in the Deutsche Bank Research Report is cumulative revenue for the full year of 2027. The two are of a different caliber.
In terms of stock price trends, SpaceX rebounded from $123.62 on November 5 to $143.09 on November 10, with a cumulative increase of about 15.75% over the three trading days; based on the latest closing price, Deutsche Bank's latest bullish research report firmly sets a target price of $250 corresponding to a huge potential increase of about 74.7%.
Deutsche Bank is particularly optimistic about the pricing advantage of computing power already in production. It believes that if construction of new data centers slows down, existing facilities such as Colossus 2 are expected to maintain high rents for a longer period of time. Whether this window can be converted into shareholder returns depends on whether the new GPUs can be powered up and delivered on time, whether customers can renew their contracts, and whether repayments can cover construction and financing expenses.
Deutsche Bank expects SpaceX's capital expenditure in 2027 to be approximately US$299.5 billion, equivalent to 1.55 times its full-year revenue forecast. Musk's projected $500 billion is the annualized revenue operating rate at the end of 2027, while Starlink terrestrial 5G and orbital computing power provide longer-term growth opportunities. The $300 target price proposed by Morgan Stanley takes into account the combined value of launch, communications, and computing power, while the “big short” Bury's valuation question falls on how much the market paid a premium for these unrealized gains. If the sell-off in technology stocks pushes up the correlation of individual stocks, closing crowded decentralized trades may also amplify short-term fluctuations. The operating indicators that test the target price of $250 are still computing power delivery, rent renewal, and cash flow after deducting capital expenses.
From rockets to GPU leasing, SpaceX's biggest business this quarter is probably not a rocket
When you think of SpaceX, what comes to investors' minds is probably a rocket. However, this quarter, more revenue may come from a business that is far less impressive: AI computing chips or large-scale AI computing power cluster leasing business for Google and Anthropic.
SpaceX leases AI computing power to companies such as Google and Reflection AI through its Colossus 2 data center. Wall Street refers to this type of business as a new type of cloud service business. According to Deutsche Bank's forecast report, this AI computing power leasing business will contribute more than half of SpaceX's revenue in the fourth quarter.
The report was released after Elon Musk attended the Barron Investment Conference on November 6. According to the report, Musk hinted that fourth-quarter revenue may increase by about 70% compared to the third quarter. Analyst Edison Yu raised the revenue forecast from US$23.6 billion to US$24.5 billion, which is higher than Bloomberg's agreed market estimate of US$22.6 billion.
According to the report, Musk also said “maybe” a “similar situation” will occur in the first quarter of 2027. If revenue increases 70% again, quarterly revenue will exceed $40 billion, far higher than SpaceX's own guidance of $33 billion to $37 billion.
Yu estimates that SpaceX's annualized revenue operating rate at the end of 2026 will reach about 104 billion US dollars. He also expects gross margin for the fourth quarter to be close to 60%, higher than the approximately 55% previously indicated by management.
The stock market can be described as responding positively to this news. After falling nearly 20% the month before, SpaceX shares rose 8.6% to close at $134.27 on November 6.
The most notable detail in the report is a table showing SpaceX's AI computing power contracts. Deutsche Bank analyst Yu estimates that SpaceX's first major customer paid an average price of $7.78 per GPU hour for different chips. According to the bank's estimates, Google paid around $11.46 per GPU hour using Nvidia's newer GB300 chip; another unnamed customer paid around $13.90 for the same hardware.
For future contracts, Deutsche Bank used a price of $14 per GPU hour in the model. By my calculations, this is about 80% higher than the mixed average price of SpaceX's first contract.
Reflection AI's agreement shows how such contracts work. According to Data Center Dynamics, the startup pays about $150 million a month to use Colossus 2's GB300 computing power; after the initial three months, either party can terminate the contract with 90 days' notice.
This withdrawal clause applies to both parties. Yu suggested that SpaceX might use this clause to renegotiate the contract price or withdraw computing power for its own AI business.
Yu wrote, “SpaceX could launch a 90-day termination clause in an existing contract and actually charge customers a higher price.” At the same time, he cautioned that Deutsche Bank did not include these “more extreme results” in the model.
After the US midterm elections, Yu also predicted that AI computing power prices will remain high for a longer period of time because he believes that the pace of data center construction will slow down, and the impact will be mainly reflected in 2028. Reduced supply means that already built computing power can get higher prices.
As a result, he raised his 2027 revenue forecast from $184.4 billion to $193.2 billion.
The variables brought about by Tesla: could the consolidated price be lower?
According to media reports, Tesla and SpaceX have set up separate committees to study potential mergers. Musk also called the idea “a good idea.” Deutsche Bank analyst Yu previously assumed that SpaceX would need to pay a higher premium to Tesla shareholders in order to obtain shareholders' vote and avoid legal disputes. This expectation may have supported Tesla's stock price.
Now, he's not that sure. “Tesla shareholders may be willing to accept lower prices or premiums to ensure the merger is completed more quickly,” Yu wrote.
Yu maintains SpaceX's “buy” rating and a target price of $250. According to my calculations, compared to the November 6 closing price, this means about 86% of potential upside; however, the stock price is still about one-third below the 52-week high of $201.80.
However, the scale of expenditure was also enormous. Deutsche Bank estimates capital expenditure in 2027 to be around US$299.5 billion, which in my calculations is equivalent to about 1.55 times its revenue forecast for that year.
Deutsche Bank is also more cautious than Musk's judgment on orbital computing power. Musk expects AI satellites to be launched in the third quarter of 2027 and expanded in 2028, but Yu believes that substantial computing power will not be achieved until 2029 or 2030.
Currently, this business story is more direct than space data centers: SpaceX is leasing AI chips at ever-increasing prices, and the leasing business is growing faster than Deutsche Bank expected just two weeks ago.
Connecting the world and renting computing power: the market dispute over SpaceX valuation lies in rent sustainability and capital return
SpaceX is trying to simultaneously establish a nationwide mobile communication network and a large-scale AI computing power platform. Qualcomm has confirmed that Starlink Mobile will use the Dragonwing platform and plans to deploy the US terrestrial 5G network starting in 2027, and build tens of thousands of sites within the multi-year agreement period.
Satellites are responsible for expanding coverage in remote areas, and terrestrial base stations take on high-capacity communication needs in densely populated areas, giving Starlink an opportunity to move from supplementary satellite services to integrated mobile communication services that compete with the three major US operators; however, national network construction and user acquisition still need to be gradually completed.
The key to the computing power business is to convert the equipment already in production into enough billable GPU hours. Deutsche Bank believes that construction restrictions may slow down new supply, thereby extending the high rent window for existing computing power. However, the $14 per GPU hour used in its future contract is a model assumption, and the average cross-chip hybrid price of $7.78 in the initial contract is not the same specification.
The two-way exit clause of the Reflection contract also means that the continuity of revenue is subject to customer renewal tests. High rents can increase the revenue potential of assets already in production, but Deutsche Bank predicts capital expenditure of up to US$299.5 billion in 2027, and final shareholder returns will still depend on delivery, repayment and financing costs.
This is also the end point of the disagreement between Damor (Morgan Stanley) and Bury, the “space head” that accurately predicted the 2007 US subprime mortgage crisis — Damo's $300 target price takes into account the synergy value of launch capacity, global connectivity, and AI computing power; Bury questioned that the market was paying too high a price for growth that has yet to be realized. *Damo's own $75 bear market scenario also includes the assumption that Starship's delay and AI deployment and monetization fall short of expectations. Looking at it from this perspective, Deutsche Bank emphasizes the revenue being realized by terrestrial computing power. Damo places more emphasis on long-term platform collaboration, and Bury's question reminds investors to test whether these benefits can cover the capital needed for expansion.