SpaceX Just Inked a New AI Deal Worth Up to $6.3 Billion

SpaceX agreed to a $6.3 billion deal with Reflection AI, which will pay $150 million per month starting July 1 for access to Nvidia GB300 chips and AI infrastructure at SpaceX’s Colossus data center. The contract gives SpaceX a new revenue stream in AI, though it includes a 90-day termination option after the first three months and comes against the backdrop of large AI losses disclosed by the company. The article frames the deal as validation of SpaceX’s AI ambitions rather than a transformational catalyst.

Analysis

The important signal is not the headline revenue but the implied willingness of frontier-model buyers to lock in scarce compute at scale before economics are proven. That matters most for NVDA: when end-customers sign capacity-style contracts, it pulls forward rack demand, supports higher utilization of next-gen GB300 supply, and reduces the market’s fear that AI capex is peaking. The second-order beneficiary is the broader AI infrastructure stack—networking, power, cooling, and data center operators—because the binding constraint is increasingly physical deployment, not model ambition.

For GOOGL, the read-through is more mixed: open-source distribution lowers the moat around closed-model APIs and can accelerate enterprise experimentation, but it also expands total AI usage and therefore indirectly validates the value of hyperscale compute. The more interesting competitive effect is on smaller model vendors: if open-source performance continues converging while access is subsidized by strategic capital, then standalone AI application companies may face margin compression sooner than the market expects. In that scenario, the winners are the picks-and-shovels suppliers rather than the pure software layer.

The key risk is that this is still venture-style economics disguised as infrastructure demand. A high monthly commitment can be renegotiated or terminated within months if model adoption disappoints, so the revenue visibility is weaker than it looks. The broader AI trade should therefore be treated as a 6-12 month theme, not a straight-line multi-year extrapolation, until these contracts survive a full product cycle and prove they convert into sustained inference and training utilization.

Contrarian view: the market may be overrating the signal value of an open-source model launch and underrating how capital-intensive it is to monetize it. Open source can drive adoption, but it also commoditizes differentiation and may force a race to the bottom on pricing. If that dynamic gains traction, the upside migrates away from model IP holders and toward the hardware layer, while many AI software names see their implied terminal margins reset lower.

Key Decisions for Investors

• Add to NVDA on any post-news consolidation; best risk/reward is a 3-6 month horizon into GB300 deployment cycles, with the thesis that capacity pre-commitments keep enterprise demand above consensus and reduce downside from any near-term digestion.
• Prefer long NVDA / short a basket of late-stage AI software names that rely on model monetization rather than infrastructure; the pair benefits if open-source adoption compresses app-layer pricing while compute spend remains durable.
• Buy GOOGL on weakness only as a secondary beneficiary, not as the primary trade; upside comes from ecosystem expansion, but the open-source trend also raises competitive pressure on proprietary model economics.
• Consider a call spread on NVDA rather than outright calls to express the view that AI capex remains bid over the next 1-2 quarters, while capping premium if the contract proves more promotional than durable.
• Avoid chasing pure-play open-source AI names after financing-driven valuation spikes; the risk/reward is poor if revenue quality is tied to short-duration capacity deals that can unwind within one quarter.