The Great Orbital Compute Pivot

NVIDIA's $20B Groq move and the race for space-based inference

December 25, 2025 Fudge Factor

Google’s Ironwood TPU v7 / v7x is 7th generation and boasts 29x power improvement over prior TPU generations. Based on the briefing I attended earlier this year on capacity engineering and road maps, recent NVIDIA and Groq news appears to be a stocking stuffer bargain even at $20B.

A fun query is to use LinkedIn people browser to marvel (no pun intended) at the number of Google hires that had NVIDIA or Groq (or both) in their prior career. You’ll find like “hardware design”, “ASIC”, “ML Accelerator”, “Validation”, and “Product Management” and can flip around past or adjacent companies too.

So, perhaps, the Groq and NVIDIA band was getting strategically poached by Google… but now the current remaining lineup is getting reformed under a drastically different investment model inside NVIDIA. Plus, the current Groq executive lineup also has past careers at NVIDIA and Google.

Orbital Compute Pivot

I believe Google’s Project Suncatcher got the attention of NVIDIA and lit a fire. The only thing more premium than a XPU on Earth is compute engineered for aerospace, orbital space, cislunar space, an outer space.

For example, Starcloud-2 will utilize NVIDIA Blackwell H100’s and B200’s for in-orbit processing of Capella Space’s SAR (synthetic-aperture radar) satellite constellation data. The real question will be what Starcloud-3 will utilize.

🚀🔭🛰️🧑‍🚀 Space Value Chain (SVC) companies and organizations

• Groq
• NVIDIA
• Starcloud

View this page on GitHub.