In the movie The Founder, Ray Kroc's accountant reveals an uncomfortable truth: “You're not in the hamburger business. You're in the real estate business.” That same logic is now being applied in the artificial intelligence industry, where SpaceX, Nvidia, and Apple have discovered that they don't need to have the best model to win; they just need to own the land, the tools, or the distribution channel.

SpaceX: The Landlord of AI

SpaceX has replicated McDonald's model: it built a massive computing infrastructure (Colossus) to train Grok, its own AI model, and now rents out that same capacity to its competitors. The result is that regardless of whether xAI succeeds or fails, SpaceX earns millions of dollars monthly from renting computing capacity.

This week, the company signed a contract with Reflection AI, a startup founded by former Google DeepMind researchers, which will pay $150 million per month to use Nvidia GB300 chips hosted in Colossus 2, the expansion of the data center in Memphis, Tennessee. If the contract holds, SpaceX could pocket around $6.3 billion. And that's just one client.

Earlier this month, Google agreed to pay SpaceX approximately $920 million per month for 32 months, totaling nearly $30 billion. Additionally, xAI reached an agreement with Anthropic that could bring up to $45 billion in revenue to SpaceX. In total, Elon Musk's company is accumulating a lease portfolio that makes it the largest owner of AI infrastructure in the world.

Like McDonald's, SpaceX uses the ultra-fast system it developed for its own restaurant (Grok) to attract tenants. The hamburger (the AI model) is secondary; the land (computing capacity) is what really matters. Elon Musk has become the Ray Kroc of artificial intelligence.

Nvidia: Mayor McCheese

Nvidia is the hardware provider everyone needs. Its GPUs and network infrastructure are present in all major AI data centers, including Colossus. The company has invested heavily in companies that rent capacity in Colossus, such as Anthropic, Reflection AI, and xAI itself. In this way, Nvidia collects a “tax” on every node of the AI economy, regardless of which model wins the race.

Colossus began in July 2024 with 100,000 Nvidia H100 Hopper GPUs and quickly doubled its capacity to 200,000 GPUs. It currently has more than 220,000 GPUs, including H100, H200, and the new Blackwell accelerators. The entire infrastructure runs on Nvidia's Ethernet Spectrum-X platform, using Spectrum SN5600 switches based on the Spectrum-4 ASIC.

Nvidia has positioned itself as the Mayor McCheese of AI: it supplies the essential tools and benefits from every transaction. Whether Anthropic, Google, Reflection AI, or xAI creates the dominant model, the computing power will have been purchased from Nvidia.

For companies looking to optimize their infrastructure, understanding how secure VPN and firewall configuration works is key to protecting their networks, especially when relying on third parties like Nvidia or SpaceX.

Apple: The One That Takes the Best Part

Apple's AI strategy is even brighter. Instead of spending tens of billions training frontier models, Apple has designed a three-tier routing system that minimizes its costs and maximizes its revenue.

When a user asks Siri something, an orchestrator integrated into the operating system decides the complexity of the task. According to third-party estimates, around 85% of requests are handled directly on the device using small, efficient models developed by Apple. These tasks include summarizing texts, prioritizing notifications, retouching photos, or suggesting replies.

Approximately 12% of queries are sent to Private Cloud Compute, Apple's own server infrastructure, which runs larger models on chips designed by the company in its own data centers. Only 3% of the most complex requests are routed to external partner models, such as Google's Gemini.

This design allows Apple to avoid the enormous cost of training a frontier model. While Microsoft, Google, Meta, and Amazon spend tens of billions of dollars annually on GPU infrastructure, energy, and research teams, Apple simply pays Google $1 billion per year to license a customized version of Gemini that powers the new Siri.

Moreover, Apple can easily switch partners thanks to the Foundation Models framework, a native Swift API with a published LanguageModel protocol that any provider can implement. Google's Gemini is compatible, and Anthropic's Claude likely will be too. Apple's orchestrator directs queries to the model that best fits that interface, so changing providers involves modifying the routing logic, not rebuilding the entire system.

Apple profits in several ways: Apple Intelligence requires recent hardware, driving device upgrade cycles, and advanced features incentivize higher-tier iCloud storage plans. While other players invest trillions of dollars accumulating debt, Apple simply collects billions without facing the massive investments needed for companies developing frontier models.

For IT professionals managing infrastructure, server virtualization with Proxmox can be an efficient alternative for optimizing resources, similar to how Apple optimizes its small on-device models.

Conclusion: AI Is Not the Product, It's the Rent

The AI industry has long told itself a story: that companies with the best models will be the winners, that intelligence is the product, and that the most capable chatbot will eventually dominate the market. That story is wrong. Some of the companies developing the best models are mere tenants. The companies that rent out computing capacity are the true owners.

Ray Kroc would immediately recognize SpaceX's strategy. The hamburger is not what matters; the land is. Right now, the most valuable land in the world is not in Silicon Valley. It's in a data center complex in Memphis packed with hundreds of thousands of GPUs. And the man who owns it has just realized that he is not in the artificial intelligence business. He is in the real estate business (and he's probably loving it).

For companies that want to understand how to protect their infrastructure in this new ecosystem, it is advisable to review success stories in enterprise network protection and the impact of government regulations on AI infrastructure.

Original source: ComputerWorld. Analysis and adaptation by ForgeNEX.