Where Will SpaceX Be in 3 Years?

SpaceX is positioning as an AI infrastructure player, with a $60B Anysphere acquisition and major neocloud contracts: Google’s three-year deal could generate about $30B for SpaceX by 2029, while Anthropic pays roughly $15B annually for three years for Colossus 1 capacity. Starlink already has 12M subscribers and $11.4B sales in 2025 with $4.4B operating income, targeting U.S. subscribers up to 15M by 2030 and a planned 1,200-satellite launch in mid-2027. Goldman Sachs estimates rocket revenue could rise to $8.3B by 2030 from $4.1B in 2025, and Starship’s expected 90% reduction in marginal launch costs could widen the competitive moat—though the article flags execution risk and urges waiting for near-term delivery.

Analysis

The real public-market winner from this narrative is not the private company itself but the infrastructure layer that gets consumed by a broader race for compute and connectivity. If even a fraction of the described AI capacity buildout is real, the incremental dollars should flow first to GPU supply and power/cooling ecosystems rather than to the story stock; that makes NVDA the cleanest listed proxy, with the best risk/reward over the next 3-12 months if capex guides stay elevated. By contrast, the “space” angle is mostly non-investable today, so public investors are likely to overtrade the headline and underappreciate how little direct revenue there is to capture in listed names.

TMUS is the most interesting second-order loser. The market may initially assign takeover optionality, but the longer-duration risk is that direct-to-device and bundled satellite broadband compresses wireless pricing power, especially in rural and low-density segments where economics are already fragile. That pressure would likely show up first in churn, then in lower net-add expectations and slower service revenue growth over 6-18 months; any M&A premium would be offset if the strategic logic is instead defensive capex or spectrum repositioning.

GOOGL is a more nuanced read: any large external compute deal is a symptom of AI demand outpacing internal capacity, which is bullish for total addressable monetization but negative for near-term margin discipline if the spend is not offset by product pricing. GS has only a peripheral fee stream unless the narrative translates into capital markets activity or M&A, so it is not the place to express this thesis. The contrarian miss is that the market will likely overweight Starship efficiency claims before they are operationally verified; until launch cadence and unit economics are demonstrated, the public-market impact should remain modest and mostly confined to sentiment around AI infrastructure and telecom substitution.

Key Decisions for Investors

• Long NVDA vs. market on a 3-12 month horizon; use pullbacks to add, with the thesis that incremental neocloud capex is still underappreciated and should support GPU demand through next earnings cycle. Falsify if hyperscaler capex guides roll over or AI server lead times normalize faster than expected.
• Short TMUS into any rally driven by takeover speculation; view it as a 6-18 month competitive-pressure trade rather than an M&A trade. Risk/reward improves if subscriber growth slows or if satellite-to-device headlines intensify; cover if a credible deal process emerges.
• Hold off on chasing SPCX-themed sentiment in public markets; there is no clean listed vehicle and the equity implications are too speculative for direct action. Treat as an alert rather than a position until there is verifiable launch/contract execution.
• Light long GOOGL only on evidence that external AI capacity is expanding monetizable usage faster than incremental compute costs; otherwise stay neutral. The right entry is after confirmation that AI revenue growth is outpacing capex intensity in the next print.
• If telecom weakness broadens, pair long NVDA / short XLC- or telecom-exposed baskets as a cleaner expression of AI infrastructure benefit versus connectivity substitution over 6-12 months. Exit the pair if wireless pricing remains stable or regulators force limitations on satellite-to-device expansion.