The Zhitong Finance App learned that as SpaceX (SPCX.US) accelerates the deployment of satellite internet and artificial intelligence (AI) infrastructure, the latest research reports from both Wall Street investment banks are optimistic about the potential return on their capital investments. Morgan Stanley predicts that by 2030, SpaceX's launch of a Starship full of 50 Starlink V3 satellites is expected to generate incremental annual revenue of about 400 million US dollars, and the cost recovery period based on incremental operating profit is only about 19 weeks. Bernstein, on the other hand, believes that if SpaceX can seize the current window period of tight supply of AI computing power and high rental prices, its terrestrial data center investment is expected to pay for itself in less than a year.
The analysis of the two investment banks together points to a key feature of SpaceX's business model: by speeding up infrastructure deployment, achieving return on investment in a short period of time, and using already built satellite networks and computing power facilities to obtain long-term benefits. However, the two agencies also emphasized that the above returns are highly dependent on factors such as Starship's reuse technology, data center construction speed, and AI computing power market price.
Morgan Stanley maintains SpaceX's “overbearing” rating and a target price of $300; Bernstein maintains a “outperforming the market” rating and a target price of $268. Based on the closing price of 143.50 US dollars on November 18, the target prices of the two institutions meant potential room for growth of about 109% and 87%, respectively.
The economic benefits of the Starship launch are highlighted: it is expected to recoup the costs in 19 weeks
Morgan Stanley pointed out in a research report released on November 18 that Starship's larger carrying capacity is expected to greatly improve the economic benefits of Starlink's satellite internet service. The bank expects that 50 third-generation Starlink satellites can be deployed in one fully loaded Starlink satellite launch, and each satellite can bring about 8 million US dollars in incremental annual revenue, for a total of about 400 million US dollars. The total communication capacity of a single launch satellite is approximately 50 Tbps, of which the capacity available for commercial sale is expected to be around 2.5 Tbps.
Meanwhile, Morgan Stanley estimates that by 2030, the variable cost of a single starship launch is about 100 million US dollars, including 60 million US dollars in launch costs and 40 million US dollars in satellite payload costs.
Based on the above assumptions, the cost of a single launch is equivalent to approximately 13 weeks of incremental revenue from the satellites it is deployed on. Based on an incremental profit margin before interest and tax of 70%, the cost recovery period corresponding to an annual incremental profit before interest and tax of approximately US$280 million is approximately 19 weeks.
This return efficiency is clearly superior to the bank's estimates of emerging cloud computing businesses. The comparison chart in the Morgan Stanley Research Report shows that according to the computing power rental price, capital expenditure, and profit margin assumptions set by the emerging cloud computing business, the payback period for revenue and profit before interest and tax is about 41 weeks and 46 weeks, respectively.
Morgan Stanley also pointed out that the downstream communication capacity of each Starlink V3 satellite is about 10 times that of existing V2 Mini satellites. Since the Starship can carry more satellites at once, the additional communication bandwidth for a single full load launch is 20 to 25 times that of a single Falcon 9 (Falcon 9) launch, which is approximately equal to the latter's Starlink satellite deployment for two to three months.
As launch frequencies increase and rocket reuse technology matures, the bank anticipates that SpaceX's internal variable payload launch cost is expected to drop from about $1,000 per kilogram in 2026 to $500 in 2030, and further drop to about $140 in 2040.
Bernstein: Seizing AI's scarce computing power window, terrestrial data centers are expected to pay back in a year
Unlike Morgan Stanley, which focuses on Starship and Starlink, the research report released by Bernstein on November 19 focused on analyzing the commercial logic of SpaceX's large-scale construction of terrestrial AI data centers.
Bernstein pointed out that Musk plans to build at least 20 gigawatts (GW) of electricity infrastructure by the end of 2027 and greatly expand the scale of computing power deployment. Although this expansion plan raised questions about investment costs and construction speed in the market, the bank believes that the current tight supply of AI computing power provides SpaceX with an opportunity to sell short-term computing power services at a higher price.
Research reports show that SpaceX has reached computing power deals at the price levels of 30 US dollars and 50 US dollars per watt this year, and the company's management believes that there is still an opportunity to obtain prices of 50 US dollars or more per watt in the future.
Bernstein believes that although this price level is difficult to maintain for a long time, it is not completely out of touch with the reality of the market. Emerging cloud computing service providers such as CoreWeave (CRWV.US), Nebius (NBIS.US), and IREN (IREN.US) all reflected strong computing power demand and pricing in the third quarter. Recently, the price range for new contracts was about $20 to $45 per watt.
The bank estimates that the construction of a 1 gigawatt computing power facility will require an investment of about 35 billion US dollars. If SpaceX can quickly sell additional computing power services at the level of $50 per watt, its investment is expected to be recovered through revenue in less than 12 months.
Bernstein emphasized that the key is not how many years the price of 50 US dollars per watt can last, but rather whether SpaceX can complete construction and lock in high-priced short-term contracts before more competitors' production capacity is put into the market.
The bank expects that as a large number of new data centers are put into operation from 2028 to 2029, the tight supply and demand situation of computing power will gradually ease, and prices will also be under downward pressure. By 2030 to 2031, the price of the original computing power service could fall back to around $13 to $15 per watt.
Therefore, for SpaceX, the speed of construction will directly determine the return on investment in the terrestrial AI business.
Ground and space collaborative development SpaceX bets on next-generation AI infrastructure
Bernstein believes that SpaceX's long-term strategy is not simply to rely on terrestrial data centers to earn computing power rental revenue, but rather to use ground facilities to complement future orbital data centers.
According to the bank's plan, SpaceX can take advantage of the high prices brought about by the lack of computing power from 2027 to 2028 to recoup its terrestrial infrastructure investment as soon as possible. After that, even if the rental price of computing power falls, the company can still use these facilities to train Grok's artificial intelligence models or continue to sell computing power to third parties at market prices.
With the gradual deployment of orbital data centers, ground facilities can also undertake tasks such as data storage, AI training, and backup, reducing customer concerns about completely relying on space computing facilities.
Bernstein predicts that SpaceX's installed terrestrial computing capacity will reach about 20 gigawatts in 2029, while the orbital computing capacity is expected to reach about 27 gigawatts in 2031, exceeding the scale of terrestrial computing power. In contrast, SpaceX's own proposed orbital computing power target for 2031 is more aggressive, reaching 77 gigawatts.
The bank pointed out that orbital data centers may reduce the pressure on land approval, power supply and cooling costs faced by traditional data centers, but their economic benefits still need to consider factors such as launch costs and equipment residual value. Unlike terrestrial data centers, the residual value of rail equipment may be close to zero after reaching the end of its useful life.
Meanwhile, Morgan Stanley's prediction that Starship launch costs will continue to drop provides another potential support for SpaceX to deploy orbital data centers in the future. If Starships can be reused at a high frequency, the deployment cost of space infrastructure is expected to be further reduced.
The speed of expansion is still the biggest test 2028 may become a key turning point
Although both investment banks are optimistic about SpaceX's long-term potential, achieving these commercial prospects still faces quite a few challenges. Bernstein pointed out that when SpaceX previously built the Colossus II data center in Memphis and South Haven, it set a record of deploying 210 megawatt facilities in 91 days, and the construction period for the second cluster was further shortened to 64 days.
However, this speed is partly due to the transformation of existing industrial facilities and the simultaneous construction of electricity supply infrastructure. As the company expands to larger new construction projects, there is still uncertainty about whether it can replicate previous construction efficiency.
Bernstein also expects that by around 2028, large US technology companies and cloud computing service providers will enter the market with a large number of additional data center production capacity. According to the bank's project reserve analysis, the scale of the relevant new facilities is close to 75 gigawatts. Even considering the project extension and actual probability of implementation, there is still about 55 gigawatts of potential additional production capacity.
This means that 2028 may become an important turning point in the relationship between supply and demand in the AI computing power market. If SpaceX fails to complete construction in time, its newly invested computing power facilities may face the risk of falling leasing prices and extended payback periods.
In contrast, the core risks that Morgan Stanley is concerned about mainly stem from Starship technology verification and commercialization progress. The forecast is that SpaceX is expected to achieve a weekly Starship launch rhythm by the end of 2027, but this goal still depends on technological breakthroughs such as the reuse of rockets and boosters and the deployment of fully loaded satellites.
The two major investment banks maintain a bullish stance with a target price of up to $300
In terms of valuation, both agencies believe SpaceX has the potential to rise further, but there are differences in their predictions for future business development.
Morgan Stanley maintains a target price of 300 US dollars. In its segmented valuation, the enterprise's AI business contributed 170 US dollars, the satellite connection business contributed 111 US dollars, X and Grok business contributed 11 US dollars, and the space business contributed 8 US dollars.
Bernstein, on the other hand, maintained a target price of $268. Although the bank raised its computing power pricing assumptions for some time to come, it also slowed down the construction progress forecast for terrestrial data centers and slightly lowered its 2027 earnings per share forecast from $1.44 to $1.43.
The joint judgment of the two investment banks is that SpaceX is building an infrastructure system that supports each other through Starship, Starlink, and AI data centers. Starships are expected to reduce space network deployment costs, while terrestrial data centers may use short-term computing power supply and demand imbalances to speed up investment recovery and support future orbital AI computing services.
However, whether it is Starship's 19-week theoretical payback cycle or terrestrial data center payback expectations of less than a year, they are all based on specific revenue, price, cost, and execution efficiency assumptions, and actual performance has yet to be verified.