SpaceX Invests 3X More on AI Than Rockets and Lost $6.3 Billion on the Segment Last Year. Should Investors Be Worried?

SpaceX disclosed $12.7 billion of AI capex last year, with first-quarter 2026 AI spending already reaching $7.7 billion and 2025 AI operating losses at $6.3 billion, far above the $657 million loss in its space segment. The company argues the investment could create an advantage in orbital AI data centers and cites a $26.5 trillion AI TAM, but the payoff remains speculative and could pressure margins ahead of its planned June 12 IPO. A $1.2 billion-per-month cloud services deal with Anthropic provides some offset, but the article frames the strategy as high-risk and capital intensive.

Analysis

The immediate market read-through is not “AI is expensive,” but that SpaceX is trying to internalize the full AI stack before the market has priced in the infrastructure moat. That favors upstream enablers more than the eventual application layer: power equipment, launch/services, thermal management, and satellite components can all see pull-forward demand if orbital compute becomes a real platform race. The first-order loser is any incumbent cloud provider assuming the data-center model remains grounded and geographically constrained; the second-order loser is capex discipline across private AI if Musk normalizes trillion-dollar TAM language and investors reward scale over margin.

The bigger competitive issue is not whether space data centers work in principle, but whether the timeline compresses enough to matter before terrestrial compute economics improve again. If the orbital thesis slips beyond 2028–2030, the investment becomes a classic conglomerate drag: massive sunk cost, limited near-term monetization, and strategic distraction from the core launch/launch-adjacent businesses that justify the premium multiple. That creates a binary setup around the IPO: near-term enthusiasm can support valuation, but any evidence that AI spend is outpacing revenue conversion will quickly re-rate the equity toward a “deep tech capex story” rather than a venture-style growth compounder.

The contrarian miss is that the cited cloud revenue offsets may not be durable moats; they are likely transitional utilization of excess compute, not proof that the orbital vision is economic. If the market begins to view the company as effectively subsidizing AI experimentation with launch cash flows, the stock could underperform peers even with strong top-line growth. Conversely, if hyperscalers validate space-based inference or edge compute, today’s spend will look prescient and the optionality could extend well beyond the headline AI TAM.

Key Decisions for Investors

• Avoid chasing the IPO on day one; wait for post-lockup/first-quarter guidance on AI capex and utilization to determine whether the business is monetizing compute or merely accumulating sunk costs.
• Relative value: long GOOGL or AMZN vs. short high-multiple private/adjacent AI infrastructure exposure if market starts pricing orbital data centers as a secular winner; both have balance sheet strength to fund optionality without forcing dilution.
• Long NVDA on a 6–12 month horizon if orbital compute rhetoric translates into sustained accelerator demand; risk/reward improves if SpaceX-like capex ambitions normalize bigger cluster orders across the ecosystem.
• Hedge the IPO with a small short-dated put spread if listed near peak enthusiasm; the setup is vulnerable to a first earnings/guidance reset once investors focus on operating losses rather than headline TAM.
• Watch TSLA as a proxy for Musk-allocation risk; if the market begins to discount strategic distraction, TSLA could underperform on a 3–6 month basis even without changes in its own fundamentals.