Elon Musk has traded luxury for a compact Airstream trailer parked beside xAI's massive Colossus computing facility in Memphis, Tennessee, where he is personally overseeing the expansion of artificial intelligence infrastructure worth an estimated $48.7 billion. The billionaire entrepreneur, who also leads Tesla and SpaceX, has been living in the trailer for more than 10 days while managing the critical final phases of bringing additional GPU (graphics processing unit) capacity online for the company's Grok AI models.

This unconventional living arrangement reflects Musk's long-standing pattern of embedding himself at production bottlenecks. During Twitter's 2022 acquisition, he slept at company headquarters. Through SpaceX's early Starship development, he lived near the Starbase facility in Texas. Now, with Colossus representing the constraint on xAI's growth, Musk has positioned himself directly at the site.

What Makes Colossus Such a Critical Project?

Colossus is not a single building but rather a sprawling two-state compute campus straddling the Tennessee-Mississippi border. The facility currently operates at approximately 1.29 gigawatts of computing power, equivalent to the electrical output of a small power plant. To put this in perspective, that's enough energy to power roughly 1 million homes, yet it is dedicated entirely to training and running artificial intelligence models.

The campus houses approximately 670,000 physical accelerators, the specialized chips that perform AI computations. The facility is divided into multiple sections, each representing different generations of GPU (graphics processing unit) technology. Colossus 1, housed in a former Electrolux appliance plant, contains 276,000 H100-equivalent GPUs and was built in just 122 days, then doubled in size within 92 days. Colossus 2, the newer expansion, is three times larger and uses exclusively Blackwell-generation chips, NVIDIA's latest and most powerful AI accelerators.

The stakes are enormous. Google has committed to renting approximately 110,000 GPUs from SpaceX under an agreement disclosed in June 2026, paying $920 million per month through June 2029. These are not research budgets; they are factory utilization numbers. Missing an energization deadline means missing an invoice.

How Is Musk Managing the Colossus Expansion?

• Civil Works and Infrastructure: The campus requires massive electrical interconnection, with SpaceX investing $35 million in the first substation and planning another $20 million for a second. The inherited grid connection was only 8 megawatts against a 340-megawatt load, necessitating substantial infrastructure upgrades.
• Liquid Cooling Systems: Colossus 1 operates 33 evaporative cooling towers with 310.4 megawatts of thermal capacity. Colossus 2 requires 217 air-cooled chillers with 412.3 megawatts of capacity and 144 condensers to manage the heat generated by thousands of GPUs running continuously.
• Power Generation Transition: The facility currently relies on 69 temporary mobile turbines, all of which must be removed by July 2027 under an agreement with the Mississippi Department of Environmental Quality signed July 30, 2026. A permanent 1.2-gigawatt gas plant is being constructed in Southaven, Mississippi to replace this temporary capacity.
• GPU Deployment Timeline: Colossus 1 contains 150,000 H100 GPUs from 2024, 50,000 H200s that arrived by February 17, 2025, and 30,000 B200s that came online July 22, 2025. Colossus 2 currently holds 110,000 B200s and 330,000 B300s, with projections to reach 612,200 B300s by Q1 2027.

SpaceX President and Chief Operating Officer Gwynne Shotwell confirmed Musk's hands-on involvement during an appearance on the All-In podcast, noting that he has been helping with construction work and remains focused on bringing additional GPUs online. Shotwell described Musk's tendency to stay close to major production projects as a defining characteristic of his management style.

Why Does This Matter for the AI Industry?

The Colossus facility represents a watershed moment in AI infrastructure. The campus is not just training xAI's Grok models; it is also serving as a rental platform for other major AI companies. Anthropic, Google, and the coding tool Cursor all have tenants running workloads on Colossus infrastructure. This shift from private research rig to commercial compute provider means that missed deadlines have immediate financial consequences.

The technical specifications underscore the scale of the challenge. Colossus 2's Blackwell GPUs are configured in GB200 NVL72 racks, which pair 72 Blackwell GPUs with 36 Grace CPUs in a single liquid-cooled cabinet. Each rack consumes approximately 120 kilowatts of power. A 210-megawatt first wave of 110,000 GPUs translates to roughly 1,530 racks, each requiring precise electrical, cooling, and networking integration.

The broader context is equally significant. xAI was merged into SpaceX in early 2026, folding the AI lab into an operating arm branded SpaceXAI. At the time of the merger, xAI alone was valued at approximately $250 billion. The Colossus facility, with its estimated $48.7 billion capital stack, represents one of the largest infrastructure investments in AI history.

Musk's decision to live in an Airstream trailer rather than commute from a conventional residence signals the intensity of the next 90 days. The timeline couples civil works, high-voltage interconnection, liquid-cooling commissioning, and rack bring-up into a single compressed schedule. For a company generating $920 million monthly in GPU rental revenue, every day of delay carries substantial cost.

The facility itself has become a landmark visible from aircraft. One warehouse roof is painted with the word "MACROHARD" in letters large enough to read from a light aircraft, a playful reference to Microsoft that underscores the competitive intensity of the AI infrastructure race.