SpaceX is poised for a big comeback after recent struggles, says TD Cowen

TD Cowen initiated SpaceX with a buy rating and $200 price target, implying 37% upside from its $145.47 Monday close; the broader analyst average target implies 57% upside. The firm expects terrestrial AI-compute leasing, including business from Alphabet and Anthropic, to become SpaceX's largest revenue stream by Q1 2027, with estimated 4Q 2026 revenue of about $8.1B and exit ARR of $41B. TD Cowen projects compute capacity will rise to 6GW by end-2027 from 2.1GW in 2026, while Starship could expand longer-term launch and government-mission revenue.

Analysis

The key valuation question is whether SPCX should be underwritten as a launch/satellite asset with an AI option, or as a power-constrained GPU lessor. The latter commands a materially higher revenue multiple but also embeds datacenter-style risks: utilization, power-delivery delays, accelerator procurement, customer concentration and rapid hardware obsolescence. A quarterly revenue run-rate claim is not equivalent to durable ARR until contract tenor, take-or-pay provisions, pass-through power pricing and customer renewal behavior are disclosed.

GOOG's participation is strategically ambiguous. Leasing external capacity can relieve near-term internal compute bottlenecks and preserve capital flexibility, but sustained dependence on a third-party provider would imply that incremental AI inference costs remain elevated relative to owned TPU infrastructure; that is modestly negative for GOOG's medium-term AI margin narrative. Conversely, a migration back to internal capacity or custom silicon would expose SPCX's customer concentration sooner than consensus expects.

Near-term price action is likely driven by sell-side target convergence and post-IPO liquidity rather than fundamental earnings revision. Over the next 1-3 months, evidence of contracted capacity, delivered power and disclosed unit economics matters more than additional technical launch milestones; the market may be assigning too much credit to separate businesses without a transparent capital-allocation bridge. Over 6-18 months, the upside case requires high utilization through successive GPU generations while avoiding debt-funded infrastructure expansion that dilutes free-cash-flow conversion.

The contrarian view is that the pullback may reflect appropriate skepticism rather than a simple entry point: the implied growth path leaves little room for commissioning delays or a moderation in frontier-lab spend. A durable rerating needs independently verifiable backlog and margin data, not analyst extrapolation from capacity targets. Falsification for a constructive thesis would be delayed energized capacity, sub-85% utilization, customer concentration above 50% of compute revenue, or guidance that implies materially lower revenue per deployed megawatt.

Key Decisions for Investors

• Keep SPCX on a conditional long watch rather than chase analyst-target momentum. Initiate only after the next results disclose contracted versus merely available capacity and support an entry in the $140-$150 range; target $190-$200 over 6-12 months, with thesis invalidated by a close below $130 accompanied by a capacity or margin guidance cut.
• For existing SPCX exposure, reduce the position into a return toward the prior $220-$225 area unless disclosed backlog and lease economics improve. That zone is likely to attract supply from IPO-era holders and requires roughly a 50% advance from current levels without yet having a corresponding fundamental disclosure catalyst.
• Use GOOG as a hedge monitor, not a direct sympathy long: track AI infrastructure capex guidance and TPU availability at the next earnings release. A sharp increase in owned-capacity capex or statements favoring internal inference would weaken the external-compute durability thesis; stable capex with continued third-party leasing would be the more supportive read-through for SPCX.
• Avoid pairing SPCX long against established GPU-cloud names until contract terms and power economics are known. The apparent competitive overlap could reverse if SPCX is primarily serving a small set of capacity-constrained strategic customers rather than competing broadly for enterprise workloads.