Shares of SpaceX have fallen roughly one-third from their post-IPO peak, but Bernstein argues that the market's focus should be on the core logic driving long-term valuation—whether orbital AI data centers can ultimately be realized—rather than short-term financial results.
Bernstein released a research report on July 31, maintaining an "outperform" rating on SpaceX with a target price of $239. As of the close on July 30, SpaceX shares were trading at $112.20, representing a cumulative decline of about 34% from the post-IPO high, with the target price implying roughly 113% upside. SpaceX is scheduled to report its first-ever earnings report since its IPO after the U.S. market close on August 4 local time.
Analyst Douglas S. Harned stated that investors need not overinterpret single-quarter financial performance; what matters more is whether management can consistently demonstrate its long-term growth trajectory. Bernstein believes that whether SpaceX can eventually move toward a multi-trillion-dollar market cap hinges not on "when" orbital data centers will be achieved, but on "whether" they can be achieved.
Starship's reusability is the core of the valuation logic
In Bernstein's view, whether Starship can achieve rapid, fully reusable capability is the most critical variable in the entire investment thesis.
The report projects that by 2031, SpaceX will execute approximately 3,600 Starship launches per year, a target based on the complete reusability of both the first-stage booster and the second-stage spacecraft. Currently, SpaceX has achieved launchpad recovery for the V2 booster, but the V3 booster has not yet been validated. The 13th test flight, conducted on July 20, was generally successful, though some engines on the booster failed to reignite before splashdown.
In contrast, the second-stage spacecraft has made greater strides. After this flight, its heat shield integrity was notably superior to previous test flights. Elon Musk subsequently indicated that the 14th test flight might see the first attempt to recover the second-stage spacecraft using the launch tower's "chopstick" mechanical arms. Bernstein believes that current technical progress is in line with expectations, but there is still a long way to go before supporting large-scale, high-frequency launches. Beyond the two existing launch pads, the company has two more under construction and is in discussions with multiple state governments to add five to six more launch facilities.
Semiconductor supply and regulation remain two major real-world challenges
The report also highlights semiconductor capacity and regulatory approvals as the two most notable risks to watch.
According to Bernstein's calculations, if approximately 20 GW of orbital computing power is deployed by 2031, with each satellite consuming 120 kilowatts, this would require about 170,000 satellites, corresponding to roughly 2.8 million wafers of annual capacity, equivalent to about five dedicated wafer fabs, with total investment exceeding $160 billion. However, the firm believes that while this scale is massive, it is not unachievable, and some capacity could be provided by third-party foundries in the future.
On the regulatory front, the U.S. Federal Aviation Administration (FAA) still needs to approve each Starship launch mission individually and has not yet authorized orbital flights. The 14th test flight could be the first attempt at an orbital flight. This week, the FAA announced it would relax some environmental restrictions to support higher launch frequency, but safety assessments for airspace and other approval processes will remain in place.
Meanwhile, the U.S. Federal Communications Commission (FCC) still needs to approve SpaceX's plans for the Starlink V3 constellation and future AI data center satellites. Currently, the company has applied to deploy approximately 100,000 Starlink V3 satellites and up to about 1 million AI data center satellites, but none have received full approval. Bernstein also noted that orbital data centers could face legal issues such as data sovereignty in the future. Proponents argue that data stored in space does not fall under any country's jurisdiction, but this legal framework remains unclear.
Technical feasibility is not the biggest obstacle
Regarding the technical feasibility of orbital data centers, which has been a focus of market discussion, Bernstein believes that most issues have clear answers.
For thermal management, two analysts with physics PhDs built a thermal model and concluded that radiative cooling is sufficient to meet requirements. By adding 30 to 50 square meters of radiator panels to the roughly 30-square-meter satellite body, it can support continuous operation of over 100 kilowatts of computing power, consistent with SpaceX's previously announced AI1 satellite design parameters—120 kW average power and 110 square meters of radiator panels. For power supply, the report notes that orbital data centers will be deployed in a dawn-dusk synchronous orbit, where satellites are almost always exposed to sunlight, alleviating earlier concerns about only about 50% solar utilization. For latency, the communication delay for low Earth orbit at 600 to 800 kilometers is about 2 milliseconds, lower than the actual experience of many ground-based data center users, and has limited impact on most AI inference tasks. Bernstein expects that SpaceX will continue to conduct model training on the ground, with orbital data centers primarily handling inference computing. As for the risk of micro-meteoroid impacts, Bernstein believes this is an engineering challenge but not insurmountable. SpaceX has already accumulated experience in collision avoidance through Starlink satellite operations, and technologies such as target detection and orbital maneuvers can similarly be applied to future data center satellites.
Comments