TechFEATURE

Space-Based AI Data Centers: What SpaceX, Google, and China Have Actually Launched So Far

E
EffectStory 編輯部Editorial Team
Published · Updated
Only two orbital AI computing efforts have hardware actually flying: Starcloud's single NVIDIA H100 satellite, launched in November 2025, and China's Three-Body Computing Constellation, which put 12 satellites into orbit on May 14, 2025, carrying 5 POPS of compute and 30 terabytes of storage. SpaceX/xAI's Starmind and Google's Project Suncatcher remain unlaunched, targeting 2027 at the earliest, with no compute figures yet disclosed.

What Are SpaceX and xAI Actually Deploying — On the Ground or in Orbit?

NVIDIA announced on August 24, 2026 that SpaceXAI, the merged SpaceX and xAI entity, will adopt its Vera CPU and Vera Rubin platform to accelerate Grok's AI workloads on the groundCITE:E1. That announcement is confirmed and ground-based; the orbital piece of the plan is a separate, first-generation satellite called Starmind, which would carry a space-optimized version of the Vera Rubin NVL72 rack into orbit, but NVIDIA's own announcement included no timeline, launch date, or compute figures for itCITE:E2.

SpaceXAI president Mike Nicolls has set an external target of launching the first Starmind satellite in the fourth quarter of 2027, followed by a larger-scale expansion in 2028CITE:E3. Satellite size and power specifications differ across media reports, and prototype testing, manufacturing, and regulatory approval have not yet been completed, placing the project at an early planning stageCITE:E3.

What Is Google's Project Suncatcher Trying to Prove?

Google's Project Suncatcher is a research program modeling a constellation of 81 satellites carrying TPUs for machine learning in low Earth orbit at roughly 650 kilometersCITE:E4. Google plans to launch two prototype satellites by early 2027 to test the concept in practice, citing that solar power in orbit can be roughly 8 times stronger and nearly continuous compared with solar power on the groundCITE:E4.

How Far Ahead Is China's Three-Body Computing Constellation?

China's Three-Body Computing Constellation, part of the Star Compute program, launched its first 12 satellites on May 14, 2025, with combined onboard compute of 5 POPS and 30 terabytes of storageCITE:E5. Those 12 satellites are described as the first step toward a planned network of 2,800 satellites, which officials say will reach a total compute target of 1,000 POPS once the full constellation is completeCITE:E5.

What Has Starcloud Actually Demonstrated in Orbit?

Starcloud launched its Starcloud-1 satellite — about 60 kilograms and roughly the size of a mini-fridge — in November 2025, carrying an NVIDIA H100 GPU that it describes as the first data-center-grade GPU operating in spaceCITE:E6. Beyond that single demonstration, Starcloud has proposed a long-term vision of a 5-gigawatt orbital data center, with solar and radiator panels roughly 4 kilometers squareCITE:E6.

What Is the Gap Between the Announcements and What's Actually Flying?

The only data-center-grade hardware currently operating in orbit is Starcloud's single H100 and China's 12 satellites; every other figure in this race — Suncatcher's 81-satellite model, the 2,800-satellite constellation, and the 5-gigawatt vision — remains a plan, not a launchCITE:E7. Claims that orbital energy could cost roughly 10 times less than on the ground are industry estimates that have not been independently verified, and thermal management, radiation exposure, launch costs, and in-orbit maintenance remain unresolved challenges for every program listed hereCITE:E7.

ProgramStatusKey NumbersTarget/Launch Date
SpaceX/xAI StarmindPlannedVera Rubin NVL72-based satellite; no compute figures disclosedQ4 2027 (first satellite), 2028 (expansion)
Google Project SuncatcherResearch, prototypes planned81-satellite model; ~650 km orbit; 2 prototypes; ~8x solar advantagePrototypes by early 2027
China Three-Body Computing ConstellationPartially launched12 satellites in orbit; 5 POPS; 30 TB storage; 2,800 planned; 1,000 POPS targetLaunched May 14, 2025
StarcloudLaunched (1 satellite)~60 kg; 1x NVIDIA H100; 5 GW long-term vision; ~4 km² panels envisionedLaunched November 2025

Across these four programs, only two have hardware actually operating in orbit today — Starcloud's single H100 and China's 12-satellite network — while SpaceX/xAI's Starmind and Google's Suncatcher remain at the modeling and prototype-planning stage, with neither disclosing compute figures for their proposed satellitesCITE:E2CITE:E4. China's declared trajectory, from 12 launched satellites toward a stated 2,800-satellite, 1,000-POPS target, is currently the only roadmap in this group that combines launched hardware with an official numeric target already in motionCITE:E5, while Starcloud's 5-gigawatt vision and the industry's 10-times-cheaper energy claim remain unverified projections built on a single orbiting GPUCITE:E6CITE:E7.

📊 Evidence

FAQ

What Are SpaceX and xAI Actually Deploying — On the Ground or in Orbit?

NVIDIA announced on August 24, 2026 that SpaceXAI, the merged SpaceX and xAI entity, will adopt its Vera CPU and Vera Rubin platform to accelerate Grok's AI wor…

What Is Google's Project Suncatcher Trying to Prove?

Google's Project Suncatcher is a research program modeling a constellation of 81 satellites carrying TPUs for machine learning in low Earth orbit at roughly 650…

How Far Ahead Is China's Three-Body Computing Constellation?

China's Three-Body Computing Constellation, part of the Star Compute program, launched its first 12 satellites on May 14, 2025, with combined onboard compute of…

What Has Starcloud Actually Demonstrated in Orbit?

Starcloud launched its Starcloud-1 satellite — about 60 kilograms and roughly the size of a mini-fridge — in November 2025, carrying an NVIDIA H100 GPU that it …

📎 Sources

  1. nvidianews.nvidia.com
  2. storagereview.com
  3. channelinsider.com
  4. research.google
  5. spacenews.com
  6. blogs.nvidia.com

Related data

Author's TakeEffectStory 編輯部

The clearest signal in this race isn't the vision numbers — it's the launch manifest. Only Starcloud's single H100 and China's 12-satellite network are actually generating compute in orbit today; SpaceX/xAI's Starmind and Google's Suncatcher are both still at the point where their own announcements disclose no compute figures at all. That gap matters because orbital data centers only make economic sense if the industry's claimed 10-times-cheaper energy holds up under real thermal and radiation stress — something none of the four programs has proven at scale yet. The metric worth watching next isn't a new pledge or a bigger satellite count on paper; it's whether China's constellation actually grows past its first 12 satellites toward the stated 2,800-satellite target, since that would be the first real-world test of whether orbital compute scales the way its promoters currently claim.

E
EffectStory 編輯部Editorial Team

Related

FEATURE

Google's $120 Billion Marvell Deal, Explained: How Custom AI Chips Are Eating Into NVIDIA's Lead

The widely reported '$120 billion' Google-Marvell deal is not an order or a revenue figure — it is a cumulative custom-chip purchase threshold Marvell must hit by fiscal 2033 to fully vest warrants Google holds. Marvell's own guidance that this Google revenue won't ramp meaningfully until fiscal 2029 triggered an 8%-plus stock drop, even as Amazon, Meta, and Google all scale in-house AI silicon against an NVIDIA that still commands roughly 70% of the AI chip market.

EffectStory 編輯部 ·
FEATURE

AI Inference Prices Are Falling About 10x a Year — What the Token-Cost Data Actually Shows

Equivalent-capability AI inference prices have fallen roughly 10x per year across multiple benchmarks, with OpenAI's GPT-4o mini priced about 99% below its 2022 predecessor and DeepSeek-R1 undercutting OpenAI's o1 by roughly 27x. Yet Google now processes about 50 times more tokens monthly than a year earlier, and Gartner projects 2025 global generative-AI spending at $644 billion, up 76.4% — proof that cheaper tokens are fueling more total usage and spend, not less.

Nathan ·