Updated
Updated · Computerworld · Jun 26
SpaceX Leases GB300 Compute to Reflection AI for $150 Million a Month
Updated
Updated · Computerworld · Jun 26

SpaceX Leases GB300 Compute to Reflection AI for $150 Million a Month

2 articles · Updated · Computerworld · Jun 26

Summary

  • $150 million a month is the price Reflection AI agreed to pay SpaceX for Nvidia GB300 capacity at the Colossus 2 facility in Memphis, giving the startup access to high-end training compute.
  • If the lease runs its full term, SpaceX could collect about $6.3 billion, extending a strategy of monetizing AI infrastructure it originally built for xAI's Grok models.
  • Colossus has grown from 100,000 H100 GPUs at its 2024 launch to more than 220,000 Nvidia chips, including H100, H200 and Blackwell-class accelerators.
  • The Reflection deal follows other large compute agreements tied to the site, including a disclosed Google commitment of about $920 million monthly for 32 months and an earlier Anthropic arrangement linked to xAI.
  • The pattern underscores SpaceX's emerging role as an AI infrastructure landlord, earning from rivals and partners regardless of which model developer ultimately wins.

Insights

As AI models become more efficient, will the multi-billion dollar compute 'landlords' face a sudden market collapse?
Is Apple’s on-device AI strategy a brilliant cost-saving move or a gamble that will leave it behind?
With AI compute now national infrastructure, who truly controls the future: tech giants or the governments depending on them?