SpaceX has pulled off one of the slickest moves in tech history—and most people haven't even noticed. The company that launched rockets and built Grok has quietly positioned itself as the McDonald's of AI: a real estate empire disguised as a burger joint. According to a Computerworld analysis, SpaceX is now collecting billions in compute lease payments from multiple AI labs, regardless of which one wins the race to build the best model. The math is staggering. This week, Reflection AI—a pre-revenue startup founded by former Google DeepMind researchers—signed a deal paying SpaceX $150 million per month for access to Nvidia GB300 chips housed at Colossus 2 in Memphis, Tennessee. If the full term runs through, that's roughly $6.3 billion flowing to SpaceX as pure landlord revenue. Before you dismiss this as pocket change, consider that Reflection AI has no widely adopted frontier model yet and was reportedly raising capital at a $25 billion valuation. They're paying SpaceX more than some countries' GDPs per year just to rent GPU time. Google is writing an even bigger check. An S-1 filing revealed the tech giant agreed to pay approximately $920 million monthly for 32 months—totaling around $30 billion. And last month, SpaceX disclosed that xAI's deal with Anthropic could bring as much as $45 billion in revenue back to the company. Add it all up and we're talking about an $80+ billion compute rental empire built on infrastructure originally constructed to train Grok. Colossus wasn't always this massive. The facility went online in July 2024 with 100,000 Nvidia H100 Hopper GPUs in Supermicro liquid-cooled HGX H100 chassis. SpaceX doubled capacity within months and now operates more than 220,000 accelerators including H100s, H200s, and next-generation Blackwell-class chips. The entire fabric runs on Nvidia's Spectrum-X Ethernet platform using the Spectrum SN5600 switch built on the Spectrum-4 ASIC. This isn't just a data center—it's a GPU sovereign nation. The comparison to McDonald's founder Ray Kroc is disturbingly accurate. In "The Founder," Kroc realizes he's not in the burger business—he's in real estate. SpaceX CEO Elon Musk appears to have reached the same epiphany about AI. Whether xAI succeeds, fails, or gets disrupted by some yet-unnamed startup matters less when you've got Google and Anthropic paying monthly rent on your GPU farms.

Nvidia: The Mayor McCheese of AI

Nvidia deserves its own category in this analysis. The chip giant has positioned itself as what the article calls "Mayor McCheese"—collecting taxes at every node of the AI economy. It supplies the GPUs and networking fabric that every frontier lab must train on, then collects hardware revenue from the winner's rivals while also profiting from the winner's success through investments in Anthropic, Reflection AI, xAI, and SpaceX itself. Whether Anthropic, Google, xAI, or some future dark horse produces the dominant model, Nvidia silicon powers their machines.

Apple Plays a Different Game

Apple's strategy is arguably even more elegant than SpaceX's. The company built a three-tier routing system for Siri: roughly 85% of requests are handled on-device by Apple's own small, efficient models (text summarization, notification prioritization, photo cleanup), about 12% go to Private Cloud Compute running larger Apple silicon models in Apple-owned data centers, and only the hardest 3% get routed to external partners. This design lets Apple avoid the ruinous cost of training frontier models from scratch. While Microsoft, Google, Meta, and Amazon spend tens of billions annually on GPU clusters and research teams, Apple pays Google a mere $1 billion per year for custom Gemini access—and can swap providers anytime through its Foundation Models framework.

Key Takeaways

  • SpaceX has transitioned from AI competitor to AI landlord, collecting rent from multiple labs regardless of industry outcomes
  • The company's compute deals total potential revenues exceeding $80 billion across Reflection AI ($6.3B), Google ($30B), and xAI-Anthropic ($45B) agreements
  • Colossus 2 in Memphis now houses over 220,000 Nvidia GPUs including next-generation Blackwell accelerators
  • Nvidia operates as a double-dip winner: selling hardware to all competitors while also investing in many of them
  • Apple's three-tier inference architecture lets it collect AI revenue without training frontier models from scratch

The Bottom Line

The AI industry has been telling itself a bedtime story—that whoever builds the best model wins. SpaceX just burned that fairy tale. Ray Kroc figured out 60 years ago that the burger doesn't matter, the land does. In 2026, Musk realized the same about compute. The most valuable real estate in tech isn't on Sand Hill Road or in Silicon Valley office parks—it's a Memphis data center complex full of GPUs. And whoever owns it just discovered they're not in the AI business at all.