Elon Musk explains how xAI built its data centers so fast
Above-market pay, relocated facility components, and industrial robots helped xAI compress build timelines at its Colossus cluster in Memphis
In a post on X, Musk said xAI built its data centers quickly by paying wages above market rates, relocating parts of the facilities, and deploying industrial robots.
The three-part playbook
The first lever is pay. Offering above-market wages is the oldest trick in construction: if you want the best crews and you want them now, you outbid everyone else.
The second lever is moving pieces of the facilities around. Musk did not spell out the mechanics. The idea fits a broader pattern at xAI of doing work wherever it is fastest and bringing the results on site.
The third lever is automation. Industrial robots do not take shifts off, and they do not need to be recruited away from competing job sites.
Colossus by the numbers
The results are easiest to see at the Colossus cluster, which sits in the Memphis, Tennessee, and Southaven, Mississippi, region. xAI has since been folded into SpaceXAI following a 2026 acquisition.
Colossus 1 was carved out of a defunct Electrolux manufacturing site. It went live in July 2024 after a build of just 122 days, launching with roughly 100,000 NVIDIA H100 GPUs.
Colossus 2 moved even faster on some metrics. After xAI acquired the site in March 2025, the facility reached approximately 200 MW of cooling capacity within six months.
By September 2026, Colossus 2 housed 110,000 GB200 chips and 440,000 GB300 chips. Musk plans to add three more batches of 220,000 GB300 units, which could push the cluster past 1.2 million advanced NVIDIA processors by the end of the year.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
More than $35.8 billion had been invested into the Colossus cluster by mid-2026.
Power first, permits later
xAI’s answer has been to bring its own power. The company currently draws hundreds of megawatts from a fleet of rented mobile gas turbines. It plans to build a permanent 1.2 GW natural gas plant.
Temporary power and cooling setups, combined with retrofitting existing industrial buildings and pre-assembling components off site, let xAI sidestep the lengthy permitting processes and interconnection delays that slow most projects.
Local critics have raised concerns about air quality, noise, and the use of unpermitted turbines. xAI has responded with removal timelines that stretch into 2027.
Renting out the firepower
xAI signed a multi-year rental agreement with Google worth $920 million per month. The deal covers about 110,000 GPUs and runs from October 2026 to June 2029.
What this means
If xAI can stand up a major cluster in 122 days, every other AI lab and hyperscaler gets measured against that benchmark.
The $35.8 billion already committed signals a long-term bet that demand for AI compute will keep climbing. The Google contract offers evidence that the cluster can generate revenue beyond xAI’s own models.
A single campus targeting more than 1.2 million GB200 and GB300 chips represents a staggering order book from one customer for NVIDIA.
The power strategy leans heavily on rented turbines and a gas plant that has not been built yet. Any regulatory crackdown on unpermitted equipment, or delays on the permanent plant, could slow the project.
What to watch next: whether xAI delivers the remaining three batches of 220,000 GB300 units on schedule, how the 1.2 GW gas plant progresses, and whether the turbine removal timelines into 2027 hold.