SpaceX and NVIDIA's orbital data center: what was announced, and who else is building compute in orbit
A graphic going around social media says SpaceX "just revealed" it will build space-based data centers powered by NVIDIA hardware. The core claim checks out, but the news is not new: it was announced in two steps in August 2026. Here is what the sources actually say, and where the project sits among everything else being built for compute in orbit.
What was announced, and when
- 4 August 2026: SpaceX said it is partnering with NVIDIA to design the compute payload for Starmind AI1, its first-generation AI satellite, with NVIDIA Rubin GPUs and Vera CPUs (Interesting Engineering). On the Q2 earnings call the same day, Elon Musk said SpaceX is "exclusive to Nvidia" (Yahoo Finance).
- 24 August 2026: NVIDIA's own press release says the first-generation Starmind AI satellite "will be based on the optimized NVIDIA Vera Rubin NVL72 rack-scale system" (NVIDIA Newsroom). The release gives no launch date. Musk said the same day that the space-optimized NVL72 system is planned for launch to orbit in Q4 2027, with "significant scale" in 2028 (Light Reading).
The satellite itself was shown in June 2026: 150 kW peak and about 120 kW average compute, a 70 m wingspan and a 20 m deployed height, with a deployable liquid radiator (Yahoo Finance). The wider plan is SpaceX's FCC application, filed 30 Jan 2026, for up to 1,000,000 satellites at 500 to 2,000 km (The Register).
What has not been published: a launch vehicle, an orbit for AI1, a satellite count for the first batch, or a cost. The tracker leaves those fields empty rather than guessing. Timelines have also moved: the launch target was "next year" on 4 August and Q4 2027 on 24 August.
Who else is putting compute in orbit
The Orbitwren tracker (data built 3 Oct 2026, 01:16 UTC) lists 17 projects from 13 companies. Five have compute hardware in orbit today:
| Project | Hardware | Launched |
|---|---|---|
| Google Project Suncatcher prototype (with Planet) | 4 Trillium-generation TPUs | 1 Oct 2026, Transporter-18 |
| TakeMe2Space MOI-1a | 117 TOPS on-board compute, 2 TB storage | 1 Oct 2026, Transporter-18 |
| Axiom Space orbital data center nodes | off-the-shelf compute, 2.5 Gbps optical links | 11 Jan 2026 |
| Starcloud-1 | 1 NVIDIA H100 GPU | 2 Nov 2025 |
| ADA Space and Zhejiang Lab, Three-Body Computing Constellation | on-board AI processors, 12 satellites | 14 May 2025 |
Google says its prototype is in contact and "operating as expected" (blog.google). Three companies have asked the FCC for large constellations: SpaceX (up to 1,000,000 satellites), Orbital (up to 100,000, SpaceNews) and Blue Origin (up to 51,600). Starcloud has filed for up to 88,000.
Try it: every tracked orbital-compute project as JSON (free, no key)
The next dated targets
None of these is on the Launch Library 2 schedule yet, so they are company targets, not booked launches:
- Starcloud-2: later in 2026 (company, March 2026).
- Crusoe Cloud on a Starcloud satellite: late 2026, GPU capacity from orbit by early 2027.
- Aetherflux Galactic Brain: first orbital data center satellite, Q1 2027.
- Google and Planet: two Suncatcher satellites with cross-links, early 2027.
- SpaceX Starmind AI1 on NVIDIA Vera Rubin NVL72: Q4 2027.
When any of them gets a launch on the schedule, the tracker flags it, starts a countdown and records the change in its daily log.
Getting the data
The free tracker page shows every project with its sources, and the full table is a free CSV on Hugging Face. The Orbitwren API returns the same rows as JSON with filters (company, status, inOrbit), the next 200 launches with rideshare and compute flags, orbits for the tracked satellites and a change log, also as a remote MCP server for AI agents:
curl "https://orbitwren.cybermaxtools.com/api/projects?company=spacex"
Independent tracker by CyberMax. Company and project names are used only to identify what is tracked; no affiliation with or endorsement by SpaceX, NVIDIA, Google or any other company listed.
