Musk Is Sleeping in a Trailer to Get xAI's Next Gigawatt Online — What That Tells the AI Compute Trade

Musk has used the same physical cue twice in his career: during Model 3 production hell at the Fremont factory in 2017–2018, and now at xAI's Memphis campus. The pattern matters because the buildout under his trailer is not a single building but a four-site cluster that spans Tennessee and Mississippi.

Two things make this expansion categorically different from the original Colossus build. First, power: at ~2 GW of IT load, the cluster consumes the equivalent of two large nuclear reactors — and because the Tennessee Valley Authority grid could not deliver that fast enough, xAI bought35 used GE LM6000-class turbines off eBay, ran them without air permits, and now faces NAACP and SELC lawsuits plus an EPA enforcement posture. Second, customer mix: the original Colossus was built to train Grok; Colossus 1 is now fully leased to Amazon-backed Anthropic for Claude inference, and SpaceX has signed Alphabet to a 32-month,110,000-GPU agreement starting October 2026.

Stacked together, the Anthropic and Google contracts alone generate roughly $26B of annualized revenue for the SpaceX compute business — before a single Grok training run is booked. That revenue line is the single biggest reason Musk can absorb capex of this scale without diluting outside shareholders, and it is also why a project that lost an estimated $5B in 2025 now shows up as a $26B-a-year gross-revenue line item in 2027 and 2028.

Every layer of the Memphis build maps cleanly onto a U.S.-listed supplier. At the chip layer, NVIDIA reported $89.0B of data-center revenue in Q2 FY2027 (ended July 26, 2026), up 117% year-over-year, and guided Q3 to $108B with the next-generation Vera Rubin platform already contributing ~20% of data-center revenue. At the rack level, Super Micro Computer has confirmed it is building the Texas gigawatt-scale site that mirrors the Memphis playbook; Super Micro's server systems are the literal enclosure for those555,000 GPUs.

Memphis is not just a build site — it is the most contested AI-infrastructure site in the United States. As of July 2026, Mississippi regulators counted 69 temporary gas turbines at the Southaven site alone, more than double the 27 turbines xAI had acknowledged when the NAACP and SELC filed their Clean Air Act notice of intent to sue in February 2026. Reuters reported in July that the turbines' pollution falls disproportionately on historically Black neighborhoods, and an SELC study in February 2026 estimated the proposed gas plant could cause millions of dollars in annual health damages across the region.

Short-term (days to quarters): the catalysts investors should track are not earnings dates but operational ones. (i) The October 2026 start date of the Google compute contract — SpaceX has a one-month grace period to deliver the 110,000 GPUs before Alphabet can terminate for non-performance. (ii) Any movement in the SELC / NAACP litigation, which could idle part of the turbine fleet and cap Memphis capacity. (iii) Nvidia's Q3 FY2027 print in late November, which is the cleanest read on whether Blackwell and Vera Rubin allocations are still going to a handful of anchor customers like SpaceX.

Long-term (one to three years): the structural question is whether the SpaceX compute business becomes a permanent, separately financeable utility — akin to a merchant power plant — or whether it stays an internal Musk asset that subsidizes Grok training. The Intel 14A Terafab deal and the planned Texas gigawatt-scale site with Super Micro both point to internal silicon and internal supply, which compresses supplier margins over time. The bull case for Nvidia is that even with internal silicon at Terafab, Musk still buys every leading-edge Blackwell generation externally for the next two cycles; the bear case is that the relationship becomes more arm's-length as SpaceX builds its own.