AI datacenters in space and the growth of space companies
Let’s start with a baseline… What are the most interesting space companies and what could their revenues look like over the next few years? Everyone thinks of SpaceX and it’s true that’s for a good reason. Next year they announced they will IPO so it’s fair to say they will continue to dominate. But let’s just get a look at the landscape.
| Company | 2025 revenue (USD B) |
|---|---|
| Boeing (Defense, Space & Security segment) | 26.42 |
| SpaceX (total) | 15.50 |
| Lockheed Martin (Space segment) | 13.16 |
| Airbus (Defence and Space segment) | 13.00 |
| Northrop Grumman (Space Systems segment) | 10.55 |
| L3Harris (Space & Airborne Systems segment) | 6.94 |
| Viasat (total) | 4.50 |
| SES (total, incl Intelsat consolidation effects) | 2.90 |
| Eutelsat Group (total) | 1.37 |
| MDA Space (total) | 1.12 |
| Iridium (total) | 0.88 |
| Rocket Lab (total) | 0.60 |
| Blue Origin (estimate; not disclosed) | 0.50 |
| York Space Systems (filing) | 0.38 |
| Redwire (total) | 0.33 |
| Planet Labs (total) | 0.30 |
But what about AI datacenters in space?
One thing that happened recently is everyone (by which I mean Musk, Jensen Huang, Sundar Pichai, Bezos, and others) started saying things like “What if we put AI data centers in space”.
The company Starcloud has a white paper explaining basically that:
- you can use a sun-synchronous orbit to always point panels at the sun continuously with no interruptions.
- you can radiate off heat on the back side of the panels and it’s not as bad as you think.
- you can get really cheap mass to space when SpaceX Starship scales (potentially, Starcloud says, $50/kg).
- if you do this at scale the cost curves will change the math on all of this to make it cheap (ish).
So what if this actually happened and people started to put data centers in space? Let’s project out some scenarios considering we already expect growth in space!
Modeling out space growth with AI data centers in space:
| Company | 2025 revenue (USD B) | Base CAGR to 2030 | 2030E base | 2030E + Space DC (moderate) | 2030E + Space DC (bull) |
|---|---|---|---|---|---|
| Boeing (Defense, Space & Security segment) | 26.42 | 3% | 30.63 | 31.63 | 33.63 |
| SpaceX (total) | 15.50 | 20% | 38.57 | 46.47 | 70.57 |
| Lockheed Martin (Space segment) | 13.16 | 4% | 16.01 | 17.51 | 21.51 (SEC) |
| Airbus (Defence and Space segment) | 13.00 | 4% | 15.82 | 16.82 | 18.82 |
| Northrop Grumman (Space Systems segment) | 10.55 | 4% | 12.83 | 14.33 | 18.33 |
| L3Harris (Space & Airborne Systems segment) | 6.94 | 4% | 8.45 | 9.05 | 10.45 |
| Viasat (total) | 4.50 | 6% | 6.02 | 7.02 | 9.02 |
| SES (total, incl Intelsat consolidation effects) | 2.90 | 5% | 3.70 | 4.90 | 7.70 |
| Eutelsat Group (total) | 1.37 | 3% | 1.59 | 1.99 | 3.59 |
| MDA Space (total) | 1.12 | 15% | 2.25 | 2.95 | 4.75 |
| Iridium (total) | 0.88 | 5% | 1.12 | 1.42 | 2.12 |
| Rocket Lab (total) | 0.60 | 35% | 2.68 | 3.68 | 7.18 |
| Blue Origin (estimate; not disclosed) | 0.50 | 25% | 1.53 | 3.53 | 13.53 |
| York Space Systems (filing) | 0.38 | 25% | 1.14 | 1.24 | 1.54 |
| Redwire (total) | 0.33 | 20% | 0.82 | 1.62 | 3.82 |
| Planet Labs (total) | 0.30 | 20% | 0.74 | 0.94 | 1.34 |
| AST SpaceMobile (total) | 0.06 | 120% | 3.30 | 3.60 | 4.30 |
| Amazon Project Kuiper (division) | 0.00 | pre-rev (modeled ramp) | 8.00 | 11.50 | 23.00 |
Breakdowns:
- Launch + on-orbit logistics (SpaceX, Blue Origin, Rocket Lab)
- Space hardware + integration (Lockheed, Northrop, Airbus D&S, Boeing BDS, MDA, Redwire, York)
- High-throughput networking + relay (SES, Viasat, Eutelsat, Iridium, Kuiper)
Comment from GPT-5.2 PRO:
Moderate vs bull is basically “a few early commercial clusters by 2030” vs “credible scaling by 2030”.
Who in the public markets gets the biggest uplift (if space datacenters work)
| Rank | Company | 2030 uplift % (moderate) | 2030 uplift % (bull) |
|---|---|---|---|
| 1 | Blue Origin | +130.7% | +784.3% |
| 2 | Redwire | +97.6% | +365.9% |
| 3 | Amazon Kuiper | +43.8% | +187.5% |
| 4 | Rocket Lab | +37.3% | +167.9% |
| 5 | SES | +32.4% | +108.1% |
| 6 | MDA Space | +31.1% | +111.1% |
| 7 | Planet Labs | +27.0% | +81.1% |
| 8 | Iridium | +26.8% | +89.3% |
| 9 | Eutelsat Group | +25.2% | +125.8% |
| 10 | SpaceX | +20.5% | +83.0% |
GPT-5.2 Pro’s explanation:
Why this ordering (mechanically): the “Space DC” wedge in the table mostly hits (1) launch + on-orbit logistics (Blue Origin, Rocket Lab, SpaceX), (2) on-orbit manufacturing/space infrastructure (Redwire, MDA), and (3) high-throughput comms networks (Kuiper, SES, Eutelsat, Iridium). Biggest % jumps go to the smaller-base companies where that wedge is large relative to their baseline.
For what it’s worth, if you ask any ai model this you get something similar to “Redwire will benefit the most, because of their Roll-Out Solar Array (ROSA) products”, and also they will say Rocket Lab because of several subsidiaries, one of which makes solar panels for satelites.
Is it actually possible, or realistic to do this now or in the future?
Let’s state some assumptions:
- A terrestrial data center is about 50 billion / GW all-in cost. (we see various sources say between 35 billion and 60 billion so let’s say 50!)
- 1 GW / year of energy costs about $200 to $250 million per year in Texas.
- Radiators are probably one of the heaviest components right now. Starcloud estimates 633 W/m² net radiator at 20°C, and very lightweight radiators are around ≤ 3 kg/m². Starcloud says the weight will be lower.
- Assume 1 kg radiation shielding per kW (Starcloud).
- Terrestrial NVIDIA GB200 NVL72 rack system is on the order of 3,245 lb fully loaded, but again Starcloud says they will get the weight down.
The summary of the weight conclusion in the white paper is:
The size of each container is limited only by ground test facilities and the payload capabilities of the next generation of heavy-lift launch vehicles, effectively capping each container at ~100 tons. The volume of the payload bay of these vehicles can accommodate ~300 racks at 50% capacity, with the remaining volume housing supporting systems. Assuming a power density of 120 kW per rack, equivalent to the NVIDIA GB200 NVL72, one launch can deploy ~40 MW of compute with rack-level mass savings.
- So: 40 MW of compute all in is 1 Starship launch. I.e. 100 tons = 40 MW = 1 Starship launch, for simplicity.
For our purposes we will ignore the details and do the following:
- Assume the MW / mass numbers Starcloud gave above (40 MW is 100 tons).
- Do a couple of multipliers e.g. 2x, 5x, 10x and compute worse mass numbers since we know mass is pretty variable!
- Compute this out for 1 GW in all scenarios so it’s apples-to-apples (somewhat!)
Current & Future potential payload costs.
| Option | Payload to LEO | Price | $ / kg | Launches to loft ~175t | Launch cost |
|---|---|---|---|---|---|
| Falcon Heavy | 63,800 kg | $97M | $1,520 | 3 | ~$291M |
| Falcon 9 | 22,000 kg | $69.85M | $3,175 | 8 | ~$559M |
| New Glenn | 45,000 kg | ~$68M | $1,511 | 4 | ~$272M |
| Starship (High) | 100,000 kg | ~$90M | $900 | 2 | ~$180M |
| Starship (Mid) | 150,000 kg | ~$30M | $200 | 2 | ~$60M |
| Starship (Low/Mature) | 200,000 kg | ~$10M | $50 | 1 | ~$10M |
Notes:
- New Glenn: $68M estimate from competitor analysis; 45t is design target (currently ~25t as they iterate)
- Starship High: Early operational phase, limited reuse, conservative payload estimates
- Starship Mid: Operational with partial reusability (6-10 flights per vehicle)
- Starship Low: Mature ops with high reuse (20-70+ flights), approaching Musk’s $10/kg target
- Musk’s aspirational target is $10/kg ($1.5-2M/launch), which would require ~100+ flights per vehicle
Results
So what does it look like in terms of costs of different launch vehicles and weights assumed vs what we have now on the ground in Texas? Below we computed the cost to push 1 GW of data center capacity to orbit. Remember: this is vs 50 billion in spend on terrestrial 1 GW.
| Vehicle | Payload (t) | Price/launch ($M) | Launch cost @1x (2,500t) ($B) | Launch cost @2x (5,000t) ($B) | Launch cost @5x (12,500t) ($B) | Launch cost @10x (25,000t) ($B) | Launches @1x | Launches @10x |
|---|---|---|---|---|---|---|---|---|
| Falcon Heavy | 63.8 | 97.00 | 3.88 | 7.66 | 19.01 | 38.02 | 40 | 392 |
| Falcon 9 | 22.0 | 69.85 | 7.96 | 15.93 | 39.74 | 79.42 | 114 | 1137 |
| New Glenn | 45.0 | 68.00 | 3.81 | 7.62 | 18.90 | 37.81 | 56 | 556 |
| Starship (High) | 100.0 | 90.00 | 2.25 | 4.50 | 11.25 | 22.50 | 25 | 250 |
| Starship (Mid) | 150.0 | 30.00 | 0.51 | 1.02 | 2.52 | 5.01 | 17 | 167 |
| Starship (Low/Mature) | 200.0 | 10.00 | 0.13 | 0.25 | 0.63 | 1.25 | 13 | 125 |
So essentially you get this:
- If Starship gets to Mid/Low level in cost the launch cost vs the cost of the equipment is a rounding error, even if weight is a big issue and never solved (10x Starcloud estimate).
- But even if you get the high-cost Starship, New Glenn, or use Falcon Heavy, you can potentially get MWs to orbit if you can keep the weight down.
I was going to keep going with this and try to calculate all-in cost over 5 years of maintenance but I feel like it’s too hand-wavy to be worth it. The main thing here is that a lot depends on Starship getting to scale.
When will Starship get to scale? The issue now is they are still testing reusability. So it seems like that’s going well but realistically it’s doubtful we see real commercial capacity before 2027.
Conclusion
Just a fun bit of analysis, and I think we can look back on this in a few years and laugh.
The reason I wanted to do this is more trying to understand timelines. And I think while long-term it may be feasible, there are a lot of engineering challenges in the next few years to solve so this is not something that is realistically going to happen in the next few years at scale.
I do expect, just because the costs of launch are actually not dominant here, that some large-scale (1 MW+, 10 MW+?) pilots will be done.