Think AI Is Expensive? SpaceX Is Spending $100 Billion to Build a New Starbase

UBS projects hyperscalers will spend $4.1T on AI infrastructure from 2026–2028 (vs. $1.3T in prior six years), with SpaceX applying the same “spend ahead of demand” logic. SpaceX plans to invest at least $100B in Starbase Louisiana, a 125,000-acre launch complex, with construction starting in 2027 and first launch targeted for 2029—about 23x its ~$4.35B annual space-operations revenue. While Starlink is already producing $11.4B revenue and $4.4B operating income in 2025 (and $4.29B/$1.66B in 2Q26), the article flags the core risk: the $100B capex pace far exceeds current space revenue, making the valuation depend on converting the infrastructure bet into durable recurring income.

Analysis

This reads less like a single-company story and more like an early warning that AI infrastructure is shifting from growth optionality to capital discipline. When capex is front-loaded ahead of monetization, the first market casualty is free-cash-flow yield, and the last is revenue; that means the valuation compression can arrive before any visible demand slowdown. Among the named hyperscalers, MSFT is most vulnerable to multiple pressure because the market already prices it as a high-quality compounder, while AMZN carries the most operating leverage if AWS margins get diluted by underutilized capacity.

The second-order effect is that the real winners are likely one layer down the stack: power, grid, cooling, networking, and construction capacity. If this spending wave persists, the bottleneck shifts from GPU supply to electricity interconnects and permitting, which tends to favor regulated and infrastructure-heavy businesses over software-like cloud names. That also means the market may underappreciate how much of this capex is effectively a race for scarce physical inputs rather than a clean technology monopoly.

The contrarian risk is that investors may be too bearish on spend intensity itself. If AI utilization improves fast enough, these outlays can look like preemptive land-grabs rather than waste, and the multiples of AMZN/GOOG/MSFT can re-rate higher despite depressed near-term FCF. What would falsify the negative view is evidence over the next 1-2 quarters that capex growth is flattening while cloud revenue growth and margins accelerate enough to keep capex/revenue below the current stress zone.

Key Decisions for Investors

• Buy 2-3 month put spreads on MSFT into the next earnings/capex update; thesis is multiple compression from FCF dilution, with risk capped if management shows utilization inflection.
• Pair trade: short AMZN / long GOOGL over 1-3 months. AMZN has the greatest cash-flow sensitivity to incremental infrastructure spend, while GOOGL has comparatively more ad-cash buffer; stop if AWS margin stabilizes and capex intensity falls.
• If expressing the broader theme, rotate part of a long XLK book into XLU for 3-6 months. The market is likely underpricing the power-grid bottleneck created by AI build-out; falsify if AI capex slows materially or utility demand does not accelerate.
• Do not trade SPCX directly; treat it as a private-market alert. Revisit only if secondary pricing, launch cadence, or Starlink cash generation can be independently verified.