Nice idea, but way too soon. Going into space calls for a much longer lifespan for AI chips. We are not there because swapping out has alway been the first choice for chips.
Again, before any of this we need my bicycle space station spinning with a near one mile diameter. We also need setrious radiation rotection up there. The wheel can be spun up edge on to the Sun with real protection on the rim.
.This scheme looks like someone throwing it against the wall and hoping it will stick right next to my fantastic gold ming project.
SpaceX Starship is Critical for AI Data Centers in Space
https://www.nextbigfuture.com/2025/11/spacex-starship-is-critical-for-ai-data-centers-in-space.html#more-206892
Without a fully reusable Starship there will not be large scale AI Data Centers in space. SpaceX has to bring the cost of flying a SpaceX Starship below $80 million per flight.
SpaceX will likely be able to do this for their own costs in 2026 but they may not sell Starship flights at less than $80 million per flight.
The Starcloud startup has $20 million in funding. They have a 13 page white paper with their plan to scale towards a 5 gigawatt AI data center in space.
CoreWeave extended its useful AI chip life calculation from 4 to 6 years (72 months) in 2023 to boost margins. This assumes repurposing for inference/edge applications after the training phase. Most others use 2-4 year AI chip life.
Starcloud has a targeted terrestrial levelized cost of electricity (LCOE) thresholds—4 cents/kWh (baseline). Earth based AI data center power – typical capital cost (Capex) range from $800–$1,200 per kW installed. However, there have been cost increases for excessive demand for natural gas turbines.
The Starcloud cost analysis for ground AI energy is to assume grid power is available at 4 cents per kwh. They do not consider build costs for larger data centers. 100% usage would then be $140+ million and another $20 million in other costs.
Starcloud assumes that a SpaceX Starship launch can be bought for $5 million. This is a flawed assumption.
They assume 40 MW of solar power can be used in space for 10 years and that the cost can be $50 per kilowatt for space based solar. This $50 per kilowatt is very optimistic and the 10 year life is unrealistic, unless there was a service to swap out the AI GPUs to get more life from the solar and other gear.
Launches to sun-synchronous orbit (SSO) are sold by SpaceX for about~$2900/kg with Falcon 9. Customer pricing is $67-70M in 2025 for Falcon 9 launch. $2800 per kg. The cost to SpaceX of Falcon 9 launches is about $15-20 million. This means SpaceX can get the cost for a full Starship equivalent payload to $150-200 million for themselves. This is above the cost of the $80 million for five years of 40 MW AI data center on earth.
No comments:
Post a Comment