sayzardTechRadar (@techradar)
SpaceX가 xAI를 인수한 뒤 머스크가 제시한 '궤도에서 저비용 AI 연산' 타임라인은 여전히 남아있는 막대한 물류적·물리적 난관을 간과하고 있다는 지적입니다. 궤도 기반 AI 연산 실현 가능성 및 현실적 장애를 문제삼는 비판적 논평으로, 우주·컴퓨팅 융합 계획의 실무적 한계를 환기합니다.
The Humanoid Hub (@TheHumanoidHub)
SpaceX가 xAI를 인수해 수직 통합 혁신 엔진을 구축하려 한다는 발표로, '우주 기반 AI'를 통해 지상 전력·냉각 한계를 우회하고 Starship으로 100만 개 위성의 궤도 데이터센터 콘스텔레이션을 발사해 연간 100GW 규모의 추가 컴퓨트를 확보하겠다는 계획을 제시한다.
The Humanoid Hub (@TheHumanoidHub) on X
SpaceX acquired xAI to build a vertically integrated innovation engine, bypassing terrestrial power and cooling limits via "space-based AI." By using Starship to launch a one-million-satellite constellation of orbital data centers, SpaceX aims to add 100 GW of annual compute
Osma A 🇫🇮🇺🇦This proposed K2 "Giga" satellite would not have enough power to run one NVIDIA Blackwell rack while on the Sun side of an orbit. Meanwhile, it also would not be able to cool one, and the computer Itself would glitch due to ionizing radiation.
#giga #orbitalcompute
https://arstechnica.com/science/2025/12/investors-commit-quarter-billion-dollars-to-startup-designing-giga-satellites/
#giga #orbitalcompute
https://arstechnica.com/science/2025/12/investors-commit-quarter-billion-dollars-to-startup-designing-giga-satellites/
Scott Francisaaaaand here's Starcloud launching an H100 into space and running Gemma on it (!) https://www.cnbc.com/2025/12/10/nvidia-backed-starcloud-trains-first-ai-model-in-space-orbital-data-centers.html
I still think all the prior technical barriers apply, and whether this is “proof of concept" or “publicity stunt" remains to be seen (I do not think this will scale to anything like what we see terrestrially, let alone make any financial sense long-term, without some profound changes in the physics of power production and heat dissipation).
I finally got around to reading this and boy is it good:
“If we use 200kW as a baseline and assume all of that power will be fed to GPUs, we'd need a system 12.5 times bigger, i.e., roughly 531 square metres, or about 2.6 times the size of the relevant solar array. This is now going to be a very large satellite, dwarfing the ISS in area, all for the equivalent of three standard server racks on Earth.”
Why training AI in orbit is a non-starter (orbital compute is NOT going to look like the same kinds of use cases we are currently obsessed with on the ground): https://taranis.ie/datacenters-in-space-are-a-terrible-horrible-no-good-idea/ #space #orbitalcompute #LEO
probably the most interesting thing I saw this week was IPNSIG Academy #25 w/Prof. Siddhartha Jayanti of Dartmouth, “On Interplanetary and Relativistic Distributed Computing” https://www.youtube.com/watch?v=IX9mVzEqzR8 #space #distributed #orbitalcompute
IPNSIG Academy: Interplanetary and Relativistic Distributed Computing
Several notable developments in orbital compute recently:
* Sophia Space is partnering with Armada to launch a fully integrated Earth-to-orbit compute infrastructure aimed at defense, energy, and autonomy use cases (and yes, AI is involved). https://sophia.space/news/armada-and-sophia-space-unveil-first-of-its-kind-fully-integrated-earth-to-space-edge-ai-platform
* Axiom Space announced plans to add optically-interconnected Orbital Data Center (ODC) infrastructure to the ISS. Two nodes will launch by end of this year. https://www.theregister.com/2025/09/17/axiom_space_aims_for_orbit/
* And my favorite is this lengthy and detailed look at the challenges and potential use cases and upsides of orbital compute from the smart cookies at Per Aspera (particularly liked that they did not shy away from problems like deorbiting risks and whether this even makes sense, although I wish they would have mentioned Kessler Syndrome and more stuff in the night sky). If you want to understand what's going on with orbital compute and why anybody thinks it makes sense to send servers into space, this is the analysis to read. https://www.peraspera.us/p/realities-of-space-based-compute
