Are data centres in space less crazy than we think?

A letters roundup covering proposals to host data centres in space, debate over AI policy in Europe, calls for taxing the rich, concerns about critical-minerals supply, notes on supermarket shopping behaviour, and commentary on China’s tech elite. Content is opinion and commentary without specific market-moving figures or actionable corporate data.

Analysis

Space-based data centres change one constraint more than proponents admit: not raw compute per dollar but location-specific latency, sovereign resilience and thermal management. With launch costs already down to the low-single-thousands $/kg to LEO and reusable vehicles on multi-year cadence, the breakeven for highly latency-sensitive edge compute shifts from decades to a 3–7 year planning horizon for specialized customers (defence, HFT, satellite-native AI inferencing). That creates second-order demand for radiation-tolerant ASICs, high-margin systems integrators, on-orbit servicing/repair and specialised insurance — not commodity hyperscaler racks.

European AI policy and higher effective taxation on top incomes tilt the competitive map toward platform and infrastructure providers able to amortise regulatory compliance and data-sovereignty costs across global customers. Expect a two-track market: large cloud providers capturing AI training loads in centralised low-tax jurisdictions, while premium regional operators and sovereign clouds capture regulated inference workloads, keeping capex for hyperscalers sticky but concentrated. Critical-minerals tightness (12–36 month cycles) will intermittently raise input costs for on-prem GPU capacity and for specialised edge hardware, accelerating reuse/recycling and long-term contracts over spot procurement.

The biggest misread is timing: space-centred compute is not a mass-replacement for terrestrial colo but a premium niche that will siphon 5–15% of the highest-value latency and sovereignty workloads within a decade, materially altering customer mix for certain REITs and specialised suppliers. Tail risks include a major launch failure, a rapid step-up in export controls or a Europe-wide AI regulation that forces data localisation and boosts demand for regional clouds instead of orbital options. Monitor launch cadence, radiation-hardened ASIC supply and multi-year procurement contracts from defence and telco customers as the earliest real-world signals of market reallocation.

Key Decisions for Investors

• Long RKLB (Rocket Lab) 12–24 month calls: exposure to launch cadence and on-orbit services; asymmetric upside if small-sat constellation servicing ramps, 3:1 upside vs time-decay if entry after next successful launch window.
• Long ALB or LIT ETF (12–36 months): play critical-minerals tightness as demand for AI edge and satellite-grade batteries grows; target 20–30% position size with stop at 15% drawdown given commodity cyclicality.
• Pair trade — short EQIX / long DLR (6–24 months): Equinix has more exposure to neutral colocation for financial services clients at risk of being replaced by premium orbital/edge options; hedge directionally with larger-cap Digital Realty which has more diversified hyperscaler and wholesale exposure.
• Long MSFT or AMZN (buy 9–18 month calls or accumulate shares): defensive play on European AI regulatory complexity and data-sovereignty spending that benefits hyperscalers who can amortise compliance costs; target 15–25% upside if enterprise AI adoption accelerates, cap losses to 12%.