Bernstein: Whether Orbital AI Data Centers Can Materialize Is Key to SpaceX Valuation
0xBroomberg
Bernstein maintains an Outperform rating on SpaceX with a $239 target, implying roughly 113% upside from the current price. This means → the bank's thesis hinges not on near-term earnings but on whether SpaceX can put data centers in orbit.
Where does SpaceX stand right now?
As of July 30, SpaceX shares closed at $112.20, down roughly 34% from the post-IPO high.
The company will report its first-ever earnings as a public company after the U.S. close on August 4.
Analyst Douglas S. Harned said investors should not over-read a single quarter; what matters is whether management can keep proving out the long-term growth path.
In plain terms = Bernstein is not grading one report card — it is asking how far SpaceX's "space story" can go.
Starship reusability — why is this the foundation of the entire thesis?
Bernstein projects SpaceX will fly roughly 3,600 Starship missions per year by 2031 — but only if both the first-stage booster and the second-stage ship achieve full, rapid reusability.
This means → if the rocket cannot land, refuel, and fly again like an aircraft, every downstream assumption about orbital compute falls apart.
Progress so far: the V2 booster has been caught at the launch pad; the V3 booster remains unverified. Flight 13 was broadly successful, but some booster engines failed to relight before splashdown.
Musk said Flight 14 may attempt the first mechanical-arm catch of the second-stage ship — this signals SpaceX is accelerating the hardest remaining step.
Semiconductors and regulation — what two hurdles stand in the way?
Chip supply: deploying roughly 20 GW of orbital compute by 2031 would require about 170,000 satellites, roughly 2.8 million wafers per year — equivalent to about five dedicated fabs and over $160 billion in investment.
In plain terms = just making the chips for space data centers would mean building a small semiconductor industry. Bernstein says the scale is large but not impossible; some capacity could come from third-party foundries.
Regulatory approvals: the FAA still approves Starship launches one at a time and has not yet cleared an orbital flight. The FCC has not fully approved SpaceX's applications to deploy roughly 100,000 Starlink V3 satellites and up to about one million AI-data-center satellites.
Orbital data centers may also face data-sovereignty questions — proponents argue data stored in space falls outside any national jurisdiction, but the legal framework remains undefined.
Does the technology actually work? Heat, power, and latency
Heat dissipation: two analysts with physics PhDs built a thermal model and concluded that adding 30–50 square meters of radiator panels to a roughly 30-square-meter satellite body can sustain 100+ kW of continuous compute — broadly matching SpaceX's published AI1 satellite specs.
Power: orbital data centers would fly in a terminator orbit — a sun-synchronous path where satellites stay in near-constant sunlight. This means → earlier concerns that solar utilization would be only about 50% do not hold.
Latency: communication delay from 600–800 km low-earth orbit is roughly 2 milliseconds, lower than many ground-based data-center users actually experience — a negligible factor for most AI inference tasks.
What is Bernstein's bottom-line call?
Technical barriers are clearing one by one. SpaceX plans to keep model training on the ground; orbital data centers would handle inference workloads.
Micrometeorite risk is an engineering challenge, but SpaceX has built collision-avoidance experience operating Starlink and can manage it through tracking and orbital maneuvers.
Bernstein's conclusion: whether SpaceX can reach a multi-trillion-dollar valuation depends not on *when* orbital data centers arrive, but on whether they can arrive at all.
This reflects a specific judgment framework — the pace of regulatory approvals and semiconductor capacity build-out will be the decisive proof points for this valuation thesis.
Content is for reference only, not financial advice.