Analyse de marché
https://www.steelldy-indices.com
SpaceX has a gross margin of 55.3%, but negative operating and net margins (-1.8% and -6.9%). Only Starlink connectivity is profitable (~+$1.66B OI), while Space and AI segments are loss-making. AWS boasts a 39.4% operating margin (+650 bps YoY), generating ~60% of Amazon’s consolidated OI from ~21% of revenue. Microsoft Cloud‘s gross margin is ~65-67% with AI pressure, but consolidated OI remains robust (+18-21%), driven by Azure and other cloud services. Google Cloud‘s operating margin reached 35.6% (vs 20.7% a year earlier), with OI tripling YoY to $8.8B. Meta faces OI pressure from AI and legal expenses, nearly eliminating FCF despite top-line growth. The key takeaway: hyperscalers are converting AI capacity into operational margins of 30-40% on cloud. SpaceX AI remains in a pure capacity leasing phase (85-90% of AI revenue) with negative consolidated profitability, positioning it closer to a neo-hyperscaler infrastructure pure-play (like CoreWeave) than mature entities like AWS or Azure.
Capital intensity and free cash flow
SpaceX exhibits extreme capital intensity with a CapEx-to-revenue ratio of 235% in Q2, far exceeding hyperscalers. Hyperscaler CapEx ratios for AI build-out in 2025–2026 typically range 70–170%, versus historical 20–40%. SpaceX’s massive CapEx is funded almost entirely by capital raises (over $100 billion post-event), while hyperscalers rely more on operating cash flow, debt, and leases, despite several experiencing temporary negative free cash flow. For instance, Amazon’s Q2 cash CapEx was $53.1 billion (TTM FCF –$7.6 billion), Alphabet’s Q2 CapEx $44.9 billion (FCF –$5.9 billion, first negative quarter in history), Microsoft’s high CapEx (~$140+ billion including leases) supported by cloud revenue ($214.4 billion), and Meta’s Q2 CapEx ~$31 billion (FCF down 91% YoY to $0.78 billion). The hyperscaler CapEx/cloud revenue ratio is at elevated levels during this AI build-out phase, contrasting with SpaceX’s outlier position and funding strategy.
Backlog, concentration, and monetization
Backlog, concentration, and monetization: SpaceX total backlog at 47.46, with 56% (~26.6) recognizable within 12 months; cloud agreements at 20.8. Critical concentration (Google + Anthropic; Colossus 1 fully leased to Anthropic). 90% of AI capacity leased to third parties, including Grok’s competitors. AWS backlog around 496 (triple-digit YoY growth); multi-year AI contracts; claimed server breakeven under 3 years. Microsoft commercial RPO at 678; ~30% recognized within 12 months. Google Cloud backlog at 514 (+50 sequential); over 50% converted within 24 months. Hyperscalers monetize through a mix of compute and software/services (Azure OpenAI, Gemini Enterprise, etc.) with sticky enterprise contracts. SpaceX AI remains dominated by raw MW leasing, without disclosure of MAU/tokens/ARPU product, making the model more sensitive to price normalization $/MW.
MSFT AAA, AMZN/GOOGL AA and Solvency
SpaceX post-fundraising: Cash & ST investments 100; Total Debt ~39.4; Equity 127; Debt/Equity ~31%; Current Ratio 5.1. Ultra-liquid but dilutive position (shares ×2 YoY).
Hyperscalers: still massive cash piles (Alphabet >100, Meta ~90, Amazon solid), moderate leverage (except Oracle), high ratings (MSFT AAA, AMZN/GOOGL AA). All are increasingly using finance/operating leases and debt for build-out, but start from a structurally positive OCF base and diversified sources (ads + cloud).
All players are in a capital intensity explosion regime driven by AI, with combined hyperscaler CapEx projected at ~$720-750 billion in 2026 versus ~$410 billion in 2025. Multi-year backlogs robustly validate demand. Free cash flow (FCF) is under temporary pressure, financed through a mix of operating cash flow and markets. SpaceX is a critical differentiator: it is 1/5 to 1/6 the size of AWS/Microsoft Cloud, with a hybrid mix of profitable Starlink, stagnant Space, and pure AI infrastructure.
Hyperscalers already have 30-40% cloud margins, while SpaceX is consolidated negative. Its capital intensity is 235% versus 70-170% for peers at peak. Monetization differs: SpaceX offers capacity leasing (including to competitors), while hyperscalers use a vertical stack (infrastructure, models, apps, ads). Funding is a massive single equity raise for SpaceX versus recurring OCF and progressive debt for hyperscalers. Under a Gordon-Shapiro model with jump-diffusion CapEx, assuming backlog conversion and yield/MW stabilization, SpaceX could achieve an “early AWS” profile by 2028-2029 if Starship scales and Grok/API monetization accelerates. In a compute price compression scenario (-30-40%), cumulative FCF for 2026-2028 becomes deeply negative, exceeding $40 billion, requiring new raises or CapEx slowdown—a more acute risk than for Microsoft (better cloud/CapEx coverage) or Amazon/Google (size and diversification).
Quantitative recommendation: SpaceX trades as a levered AI-infra and connectivity compounder at a high multiple, but with higher execution and concentration risk than established hyperscalers. In a multi-strategy portfolio, it should have a lower relative weight (0.8-1.5x vs. cloud peers), with hedges via short correlation for pure-play AI infra and puts on hyperscalers in case of CapEx cycle normalization. The expected Sharpe ratio is attractive only if backlog conversion exceeds 70% and AI margins turn positive by end of 2027.
