Big tech's next move is to put data centers in space. Can it work?

SpaceX/xAI announced a plan to deploy orbital data centers and Bloomberg reports SpaceX filed confidentially with the SEC aiming for an IPO this summer. Musk claims space-based AI could become cheaper within 2–3 years, but experts note major hurdles: matching a 100 MW terrestrial data center would require facilities 500–1,000x the International Space Station's power infrastructure, and current launch costs (~$1,000/kg) must fall to roughly $200/kg. Technical challenges include cooling (large radiators), inter-satellite laser links and latency, and heavy reliance on Starship for economics, making timelines and near-term commercial viability uncertain.

Analysis

Putting compute into orbit creates a bifurcated supply chain: one that scales mass-to-orbit and one that specializes in hardened, low-SWaP (size, weight and power) AI accelerators and optical inter-satellite fabric. The market that will pay first is not generic hyperscale workloads but customers with extreme latency, geographic resiliency, or regulatory needs; that shapes product design and means incumbent cloud vendors and GPU suppliers will sell bespoke variants rather than off-the-shelf datacenter racks.

Thermal management, repairability and launch economics are the three technical constraints that set a multi-year adoption curve: expect incremental technical demonstrations to unlock discrete capex inflection points rather than a single binary ‘go’ event. That structure creates a sequence of tradable catalysts — successful in-orbit thermal prototypes, terabit-class laser links, and concrete per-kg launch-cost reductions — each of which will re‑rate different parts of the ecosystem.

Second-order winners will be specialist subsystem suppliers (optical comms, radiator/thermal loop integrators, advanced composites and satellite on-orbit servicing) and cloud providers that convert IP into vertically integrated space-as-a-service offerings. Losers are the most complacent terrestrial infrastructure owners who assume an uninterrupted CAGR in on-prem rack demand; their valuations embed a smooth demand curve that orbital compute will puncture if the above technical thresholds are met within 3–7 years.

Key Decisions for Investors

• Long NVDA via a defined-risk call spread (buy Jan-2028 calls, sell higher-strike Jan-2028 calls) sized for 2–3% portfolio exposure. Rationale: terrestrial AI demand remains dominant over the next 12–24 months while NVDA benefits from custom accelerator variants for space prototypes; target 2.0–3.0x payoff if AI spending stays robust, stop-loss = 50% premium loss.
• Overweight GOOGL (GOOGL) equity + buy 18–30 month calls (25–50bp active position). Rationale: optionality on proprietary orbital projects and vertical integration to capture space-network services; asymmetric upside if prototypes validate cluster operations, downsize position if regulatory headwinds emerge.
• Tactical short Digital Realty (DLR) 6–24 month exposure (options or equity hedge) sized to 1–2% notional. Rationale: valuation assumes uninterrupted rack growth; downside risk if large customers divert incremental high-intensity workloads to alternative architectures (on-orbit or edge) over a multi-year window. Hedge with sector-long position (e.g., Core infra names) to limit macro beta.
• Long selective launch/subsystem suppliers (e.g., public small-launch or optical-comm suppliers) with tight sizing (1% portfolio each). Rationale: these firms win if frequent, lower-cost lifts and lasers become industry primitives; binary delivery risk is high, trim into successful milestone deliveries (first high-throughput demo, first commercial slipstream).