Musk goes exclusive with Nvidia on SpaceX earnings call, citing Vera Rubin architecture
SpaceX's second-quarter earnings call put $NVDA in focus after Elon Musk stated the company would build exclusively on Nvidia, naming the Vera Rubin architecture as the reason. The declaration places SpaceX in a…
Key takeaways
- Elon Musk stated on SpaceX's second-quarter earnings call that the company would build exclusively on Nvidia, citing the Vera Rubin architecture.
- Nvidia reported second-quarter data center revenue of $89 billion (up 117% year over year) versus AMD's $6.7 billion (up 107%), with market caps of roughly $5.3 trillion and $830 billion respectively.
- Nvidia trades at a forward price-to-earnings ratio of around 23, while AMD trades at roughly 67 times forward earnings.
- SpaceX leases its Nvidia GPU capacity, with deals including Anthropic at $1.25 billion per month through 2029, Google Cloud at $920 million per month through mid-2029, and Reflection AI worth up to $6.3 billion for GB300 chips.
- Major hyperscalers such as Amazon, Alphabet, Microsoft, Oracle, Meta, Anthropic, and OpenAI continue dual-sourcing from both Nvidia and AMD.
SpaceX's second-quarter earnings call put $NVDA in focus after Elon Musk stated the company would build exclusively on Nvidia, naming the Vera Rubin architecture as the reason. The declaration places SpaceX in a different position from most hyperscalers, who have kept AMD ($AMD) in the stack alongside Nvidia because supply is tight and different workloads favor different silicon. For AMD, losing even one high-profile name to a single-vendor commitment tightens the positioning math.
The numbers behind the gap
Nvidia reported second-quarter data center revenue of $89 billion, up 117% year over year. AMD posted $6.7 billion in data center sales over the same period, a 107% increase but off a much smaller base. The scale difference shows up in market capitalization: Nvidia sits at roughly $5.3 trillion, AMD at around $830 billion.
Valuation sharpens the comparison. Nvidia trades at a forward price-to-earnings ratio of around 23. AMD trades at roughly 67 times forward earnings. Investors are paying nearly three times as much for AMD's next dollar of earnings despite Nvidia holding the larger software stack, the rack-scale systems, and now the public endorsement from the operator of two of the most closely watched AI computing programs in the world.
What the SpaceX compute business reveals
Beyond the preference statement, SpaceX has turned its Nvidia procurements into an active infrastructure leasing business. Anthropic signed a cloud services agreement with SpaceX worth $1.25 billion a month through 2029, accessing capacity across Colossus and Colossus II, two supercomputers housing roughly 325,000 Nvidia GPUs. Google Cloud committed $920 million a month for access to about 110,000 Nvidia GPUs through mid-2029. Reflection AI recently added a multiyear deal worth up to $6.3 billion for Nvidia GB300 chips at a reported $150 million per month.
SpaceX is running GPU clusters across orbital data center design, Starlink operations, Starship mission planning, and autonomy simulation work similar to what Tesla runs. The near-term item to watch is Starmind, a satellite payload built on Nvidia's Vera Rubin NVL72 system.
The hyperscalers including Amazon, Alphabet, Microsoft, Oracle, Meta Platforms, Anthropic, and OpenAI are still dual-sourcing from both Nvidia and AMD. AMD's MI300 systems remain in those stacks and the next-generation MI400 accelerators are working through design wins. The crowded side of this trade stays Nvidia, and the setup for AMD depends on whether those hyperscaler commitments hold and the MI400 cycle converts.
What to watch: AMD's MI400 design-win announcements and any new compute agreements tied to Starmind or Colossus capacity.