Musk is naming data centers after meat products and shooting ranges while fighting civil rights lawsuits over unpermitted gas turbines — somehow this is the most normal AI infrastructure story of the week.
The Summary
- Musk is expanding his Colossus 2 AI supercomputer campus in Tennessee with two facilities: a former Hormel plant ("Spam") and an ex-police gun range, both operational by year-end
- The expansion comes as a federal judge rejected SpaceX's motion to dismiss an NAACP lawsuit over unpermitted methane gas turbines powering the complex
- xAI, now owned by SpaceX, built the original Colossus supercomputer in Memphis in 2024 to power its Grok chatbot — Colossus 2 is the second iteration demanding more compute
The Signal
The race for AI compute is turning into a real estate land grab with regulatory shortcuts. Musk is converting whatever industrial space he can find near Memphis into GPU farms, because when you need exaflops of compute to train the next Grok iteration, you don't wait for purpose-built facilities. You take the Spam factory. You take the gun range. You move fast and let the lawyers clean up later.
The timeline tells the story. Original Colossus went live in Memphis in 2024. Colossus 2 construction started last year. Now Musk is doing midnight site walks and promising two more facilities operational within weeks. This is infrastructure deployment at SpaceX velocity, applied to AI. The question is whether the regulatory framework can keep pace, or whether it even should.
"Gun Range is tracking to be operational by end of November and Spam in December."
Here's what makes this more than just Musk being Musk: the NAACP lawsuit isn't about NIMBYism or noise complaints. It's about unpermitted methane gas turbines running in Southaven, Mississippi — 20 minutes from Memphis — to generate the massive power these supercomputers demand. The complaint alleges xAI is operating critical energy infrastructure without permits. Last week, a federal judge denied SpaceX's motion to dismiss. That means discovery, depositions, and a public record of exactly how these facilities are being powered and permitted.
The AI buildout has a power problem that nobody wants to talk about honestly. Training frontier models requires compute clusters that draw megawatts. Data centers are already straining regional grids. The hyperscalers — Microsoft, Google, Amazon — are signing deals with nuclear plants and restarting retired reactors. Musk's approach appears to be: bring your own methane turbines, ask forgiveness later.
Key infrastructure constraints:
- AI training runs need consistent, massive power (measured in megawatts, not kilowatts)
- Existing grids can't handle sudden data center loads without years of planning
- Permitting for new power generation takes longer than building the compute facility itself
This creates a mismatch: AI companies can rack servers in months, but getting clean, permitted power takes years. So you get creative solutions like unpermitted turbines in Mississippi. Or you do what Meta is doing and co-locate next to power plants. Or you wait, and let OpenAI or Anthropic train the next model first.
The community opposition isn't incidental either. Packed hearings, protests, petitions — residents near Memphis understand these aren't just buildings. They're industrial power facilities that happen to run matrix math instead of stamping steel. The externalities are real: grid load, emissions, noise, traffic. The benefits accrue to xAI shareholders and Grok users. The costs stay local.
The Implication
Watch how this lawsuit resolves. If xAI loses or settles with meaningful operational restrictions, it sets a precedent for how AI companies can (or can't) bootstrap their own power infrastructure. If they win or the case drags on while the turbines keep running, expect more companies to pursue the "better to beg forgiveness" model for energy.
For anyone building in AI: compute access is the new moat, and compute requires power. The companies solving power and cooling at scale will own the infrastructure layer of the agent economy. If you're betting on AI, you're also betting on energy policy, grid capacity, and whether communities will accept data centers as neighbors. Musk is stress-testing all three simultaneously.