SpaceX Stock Loses $1 Trillion: Why the Pentagon AI Deal Won't Fix the Valuation Problem
SpaceX shares closed near $124 on July 17, roughly 45% below their June 16 peak and below the company's $135 IPO price, leaving a market capitalization near $1.6 trillion. The proximate trigger was a last-second abort of a Starship test flight after Raptor engines failed to ignite, compounding weeks of profit-taking and anxiety over an August lockup expiration that will release shares equal to 143%-214% of the IPO float. News that SpaceX is in early talks with the Pentagon to sell several billion dollars of AI data-center capacity produced only a brief intraday bounce before the stock resumed falling—a telling signal that good news no longer automatically expands the multiple.
The Real Story Beneath the Headline
The trillion-dollar "loss" is partly an optical artifact. SpaceX sold roughly 639 million shares in its IPO against 13.2 billion shares outstanding—under 5% of equity. A peak valuation near $2.66 trillion was extrapolated from trading in a thin float; the current correction is less a verdict on fundamentals than the predictable transition from scarcity pricing to genuine price discovery as insider shares unlock.
Compounding this, the IPO doubled as a disguised recapitalization: SpaceX acquired xAI in February 2026 and raised a $20 billion bridge loan to retire X and xAI debt, committing IPO proceeds to that repayment within six months. Effectively, public shareholders inherited the financing burden of Musk's weaker-margin assets.
Compute Demand Is Real—But Mismatched
The AI infrastructure narrative has genuine substance. Disclosed contracts with Anthropic (~325,000 GPUs, $1.25 billion/month), Google (~110,000 GPUs, $920 million/month from October 2026) and Reflection AI ($150 million/month) imply roughly $27.8 billion in annualized revenue—exceeding SpaceX's entire 2025 revenue base. Yet this "backlog" carries almost no duration protection: both Anthropic and Google can exit on 90 days' notice. SpaceX is financing multi-year, rapidly depreciating hardware against contracts customers can walk away from in a quarter—a profound mismatch that leaves SpaceX holding stranded-asset risk while customers retain full flexibility.
The economics already show strain. Q1 AI segment revenue was $818 million against a $2.47 billion operating loss and $7.72 billion of capital expenditure—9.4 times segment revenue. Meanwhile Starlink, the only profitable segment ($2.09 billion in Q1 adjusted EBITDA), is increasingly subsidizing Starship, AI losses and compute buildout, even as subscriber ARPU fell 22.9% year-over-year on international expansion.
The Pentagon Reality Check
Executives should resist reading the Pentagon talks as validation of SpaceX as a fourth hyperscaler. The Defense Department killed its single-vendor JEDI procurement in 2021 specifically to avoid concentration risk, and its successor architecture is deliberately multi-vendor. The Pentagon's declared end-state—a $29.5 billion "AI Arsenal" of government-owned, SCIF-accredited data centers—points toward reducing reliance on outsourced capacity, not entrenching it. SpaceX may capture overflow demand and classified workloads, but a hypothetical $5 billion award represents roughly 0.3% of current equity value. It matters strategically, not as a valuation catalyst.
A Business Model Reclassification, Not a Software Story
The essential insight for this stock is that SpaceX is migrating down the valuation hierarchy—from a perceived space-and-software monopoly toward a capital-intensive infrastructure conglomerate, and the market has not yet repriced it accordingly. At roughly 87 times 2025 revenue and 248 times adjusted EBITDA, shares still assume simultaneous success across Starship commercialization, Starlink profitability, high compute utilization and AI monetization.
The more revealing signal is what SpaceX's own contract terms disclose: renting GPUs currently has clearer economics than selling proprietary AI models, evidence the model layer remains commoditized—not proof of SpaceX's AI superiority. Aggressive pricing to undercut neoclouds benefits customers, GPU makers and power suppliers; SpaceX, as the capacity landlord, absorbs the volatility.
The base case—65% probability—sees operations improve while shareholder returns stay constrained by dilution and capital intensity, with shares more likely to trade in a $90-$150 range over 12-24 months than reclaim prior highs. A tail scenario, where a major customer exercises its termination right alongside further Starship delays and Pentagon capital redirection toward government-owned infrastructure, could compress equity value toward $600-$900 billion. The strategic conclusion: underweight SPCX equity while favoring accelerator, power and classified-systems suppliers carrying less customer-duration risk.
not investment advice