SpaceX Is Becoming An AI Giant

SpaceX is targeting 5-10 GW of compute capacity in 2027, positioning AI as potentially its largest long-term revenue engine. SemiAnalysis estimates that 10 GW could generate roughly $300 billion in annual recurring revenue if SpaceX externally monetizes half of incremental capacity. Starlink's 2025 EBITDA margin was 62.9%, although its Connectivity business required approximately $4.18 billion in capital expenditures.

Analysis

The key underwriting question is not whether SpaceX can create AI demand, but whether it can finance and power a multi-gigawatt build without diluting the cash-generative connectivity franchise. At current hyperscale benchmarks, 5-10 GW implies roughly $150-400 billion of cumulative data-center, accelerator, grid-interconnect and generation investment depending on hardware density and owned-versus-leased infrastructure. That capital intensity would likely force substantially larger external funding rounds, making valuation sensitivity to AI utilization and power procurement far more important than headline revenue potential.

The most investable near-term read-through is demand acceleration for AI electrical infrastructure rather than a direct SpaceX proxy. VRT, ETN, PWR, GEV and CEG benefit if another credible buyer enters the queue for power gear, grid upgrades, cooling and firm generation; the bottleneck is interconnection and power delivery, not accelerator availability alone. NVDA and AVGO may receive incremental demand, but a SpaceX build would be a 2027 procurement catalyst and should not materially alter their next-two-quarter earnings setups.

Consensus may be assigning too much value to a theoretical external-compute revenue stream before customer contracts, site selection, power agreements and a financing plan are visible. A 10-GW asset base is economically attractive only at sustained high utilization and premium GPU pricing; AI inference price deflation, customer concentration, or a competing hyperscaler capacity glut could sharply reduce returns. Conversely, signed long-duration power contracts and committed third-party compute customers would justify a step-up in private-market valuation well before revenues appear.

Key Decisions for Investors

• Do not add directional exposure to SPCX solely on this target; treat it as a watch item until the vehicle's actual SpaceX ownership, NAV discount/premium, liquidity and fee structure are independently verified. A funding round at a materially higher valuation without disclosed AI capex commitments would be a sentiment catalyst, not fundamental confirmation.
• Build a 6-18 month basket long VRT, ETN and PWR, sized modestly, as the higher-probability second-order beneficiaries of incremental AI campus construction. Risk/reward improves on 10-15% pullbacks; falsify if AI infrastructure order growth decelerates for two consecutive quarters or backlog conversion weakens.
• Add CEG or GEV only after evidence of contracted incremental load or power procurement emerges; grid and generation names are the cleanest expression of the binding constraint but are vulnerable to power-price normalization and permitting delays. Use a 12-24 month horizon rather than a launch-date trade.
• Monitor for four confirmation signals over the next 3-9 months: named compute customers, site/power announcements, disclosed capital partners, and accelerator supply commitments. Absence of at least two of these by mid-2027 would materially weaken the external-compute valuation thesis.