Nvidia vs. AMD: Elon Musk Picked a Side on the SpaceX Earnings Call

SpaceX said it will build its AI infrastructure exclusively on Nvidia's Vera Rubin architecture, reinforcing Nvidia's position in Musk-led operations alongside Tesla. SpaceX is monetizing Nvidia compute through major capacity agreements, including Anthropic at $1.25B per month through 2029, Google Cloud at $920M per month through mid-2029, and Reflection AI at up to $6.3B. The article favors Nvidia over AMD, citing Nvidia's $89B Q2 data-center revenue (+117% YoY), versus AMD's $6.7B (+107%), and Nvidia's lower forward P/E of roughly 23x versus AMD's 67x.

Analysis

The investable implication is less incremental GPU demand than a potential shift in the composition of demand toward fully integrated rack-scale deployments. If SpaceX’s reported leasing activity is independently confirmed, NVDA captures not only accelerator content but a larger share of networking, systems software and serviceability economics; AVGO, ANET and VRT are likely second-order beneficiaries. This would also validate a new customer category—compute owners monetizing capacity to third parties—which can extend the AI capex cycle but raises eventual counterparty and utilization risk versus hyperscaler-owned demand.

AMD’s risk is not loss of all AI share; it is becoming the price-clearing alternative for buyers seeking negotiating leverage. That outcome can preserve MI-series volumes while limiting gross-margin expansion and sustaining an earnings-multiple discount to NVDA over the next 6-18 months. The near-term catalyst is verification of capacity contracts, delivery schedules and power availability; headline contract values should be discounted until disclosed as committed take-or-pay revenue rather than contingent reservations. A reversal would come from evidence that proprietary accelerators, AMD’s next platform, or customer concentration concerns force NVDA system pricing lower.

Consensus may overread a single customer’s architectural preference as broad proof of durable exclusivity. Musk-linked demand is strategically valuable but can be volatile, internally financed and unusually exposed to execution at TSLA/SPCX. The cleaner thesis is therefore relative: NVDA benefits from system-level lock-in, while the most crowded risk is paying for perpetual scarcity after supply catches up. Monitor NVDA data-center gross margin, lead times, and receivable growth rather than announced GPU counts.

Key Decisions for Investors

• Initiate a 3-6 month long NVDA / short AMD dollar-neutral pair after confirming that the reported compute-leasing agreements are contractual and funded. Target 15-20% relative outperformance; exit if AMD discloses material MI400 hyperscaler deployments with margin-accretive pricing or NVDA guides to sequential data-center gross-margin compression.
• Add selectively to ANET and VRT on confirmation of new rack-scale buildouts, rather than chase NVDA alone. These names offer downstream exposure to Ethernet fabric and power/cooling intensity; size for a 10-15% drawdown if deployments are delayed by grid or data-center permitting constraints.
• Avoid treating SPCX-related revenue claims as an NVDA earnings catalyst until customer commitments, utilization and payment terms are independently verifiable. Set an alert for NVDA’s next earnings call: backlog conversion, networking revenue and receivables are the key validation metrics.
• For TSLA, keep AI-infrastructure enthusiasm separate from the auto earnings thesis. Any material increase in external GPU spend without corresponding robotaxi/FSD monetization would be a 6-12 month free-cash-flow headwind; reassess if capex guidance rises while automotive gross margin declines.