Orbital data centres: power constraint forces compute off-planet
Worth highlighting from today’s briefing.
- New & Emerging
Orbital data centres: power constraint forces compute off-planet
AI hyperscale demand has hit a hard ceiling: a single facility now requires 100–500 MW, and US grid capacity and permitting timelines (5–7 years) cannot keep pace. SpaceX filed FCC plans in January for up to one million data-centre satellites. Starcloud filed for 88,000. Google’s Project Suncatcher is developing radiation-hardened TPUs for orbital deployment. Nvidia’s GTC 2026 launched the Space-1 Vera Rubin Module for in-orbit AI compute; Starcloud has already trained an LLM in space on Nvidia H100s. This is no longer speculative infrastructure.
• Source: Semiconductors Insight
AI chip market: GenAI chips approaching half of all semiconductor revenue
Deloitte forecasts genAI chips near $500B revenue in 2026, roughly 50% of global chip sales. TSMC’s 3nm capacity is sold out through 2028. Nvidia faces rising competition in China — Chinese GPU makers captured 41% of AI accelerator server market in 2025. CoreWeave announced compute deals with both Anthropic and Meta this month, driving a surge in AI infrastructure stocks.
• Source: Deloitte 2026 Semiconductor Outlook
Bankwatch 