Elon Musk’s AI Empire: Colossus, Cortex & The Mega-Terafab

Kind: captions Language: en Welcome to this explainer. Today we're diving into something absolutely wild. The massive scale and the very real physical consequences of Elon Musk's newest AI superclusters. You know, we usually think of artificial intelligence as this invisible weightless thing, right? Like it's just floating out there in the ether. But today, we're looking at the staggering heavy realworld infrastructure that it actually takes to bring it to life. Okay, let's dive into this. I want you to completely rethink your concept of the cloud. We throw that word around all the time and it makes computing sound so light, fluffy, and ephemeral. But the reality of modern AI development, it's anything but. It is sprawling. It's heavy, and it is heavily industrial. What it actually takes to build the world's most powerful AI involves immense factories, millions of gallons of water, and literally enough electricity to power entire cities. To really understand the scale we're dealing with here, just look at this number. 122 days. That is the staggering speed of innovation right now. In just under four months, they built the initial Colossus Supercomputer in Memphis, Tennessee. Within that incredibly short window, they somehow brought an initial 100,000 Nvidia H100 GPUs online. The pace is absolutely relentless. Section one, inside Cortex and Colossus. Now, what's really interesting about this is the jaw-dropping volume of processing power we're talking about and how it's deployed into entirely different locations. On one side, we've got Tesla's Cortex down at their Giga Texas plant in Austin. That one boasts 70,000 AI servers featuring 50,000 Nvidia H100s and eventually 20,000 of Tesla's very own dojo units. Then, on the flip side, there's XAI's Colossus in Memphis. This beast was built inside a 785,000t former Electrolux factory. It launched with 100,000 H100s and get this, they are planning to scale it up to an astonishing 1 million GPUs. Let's look at the rapid development of Colossus. Elon Musk was originally told by experts that building a data center of this magnitude would take roughly 18 to 24 months. Instead, they found an abandoned factory and site preparation began in just 19 days. By March 2024, the site was selected. By July, Colossus 1 was operational. Fast forward to June 2025 and the facility has expanded to house 230,000 mixed GPUs featuring the absolute latest and greatest hardware. We're talking H100s, H200s, and GB20s. But why build all of this? Well, Musk has stated the goal is to solve realworld AI, but the actual applications for these two superclusters are pretty distinct. Cortex over in Austin is totally dedicated to training Tesla's full self-driving or FSD as well as their upcoming cyber taxis and those autonomous humanoid Optimus robots. Colossus, however, is busy training the XAI language model Grock, all while providing massive computing support for X, SpaceX, and even outside clients like Enthropic and Google. And the financial scale of Colossus, it's just as massive as its hardware. Google recently agreed to rent access to 110,000 GPUs for a whopping $920 million a month. Yeah, almost a billion dollars monthly. Meanwhile, Anthropic is renting out all of the compute capacity at the original Colossus 1 site. Because of this insane demand, XAI actually had to expand across the border to a second data center in South Haven, Mississippi, and they purchased a massive new 1 million square foot site in the White Haven area of Memphis for nearly $80 million. Section two, powering the supercomputers. And this brilliantly illustrates the real challenge here. We have all this unprecedented computing power, but what does it actually take to keep it running? At launch, the Cortex cluster in Texas required 130 megawatt of power and cooling. By 2026, it's projected to scale up to 500 megawatt. To put that into perspective for a second, 500 megawatt is roughly equivalent to the entire power output of an average coal power plant. Handling that kind of heat generation means building absolutely massive cooling fans just to keep the servers from literally melting down. Over in Memphis, Colossus requires an equally massive amount of energy. When its processing capability doubled, its maximum power consumption jumped from 150 megawatt straight to 250 megawatt. As you can imagine, this has placed an incredible strain on the local utility provider, Memphis Light, Gas, and Water. For context, their total peak demand is only around 3 gawatt for the entire city. And guys, it is not just electricity, it's water. Colossus requires over 5 million gallons of water every single day just to keep from overheating. 5 million. To help mitigate this staggering draw on local resources, XAI is actually building a dedicated $80 million wastewater recycling plant in collaboration with local authorities to help process water specifically for the supercomput. But before the grid could fully support Colossus, they had to improvise. The initial grid connection in South Memphis was only providing 8 megawatt. So to power the facility, XAI temporarily set up 35 methane gas turbines. These generators just steadily burned natural gas to generate about 72 megawatts of electricity. That's equivalent to roughly 3% of the entire Tennessee Valley Authority power grid, all running locally right there at the site. Section three, community and environmental impact. Naturally, a sudden massive industrial footprint appearing basically overnight has led directly to real world consequences for the community. Many residents, city council members, and even environmental agencies were completely unaware of the Colossus project until they literally saw it on the local news. Kashan Pearson, a local community leader, noted that they are experiencing more days where it is straight up unhealthy to go outside. At public hearings, residents have expressed severe frustration, citing chronic respiratory issues related to living in an already polluted section of Memphis. So, the crucial point is how this has escalated legally. In April 2026, the NAACP, teaming up with the Southern Environmental Law Center and Earth Justice, officially filed a lawsuit against the company. The lawsuit outlines allegations that the operation of what they call the Colossus Gas Plant is emitting significant amounts of harmful pollutants, including nitrogen oxides and formaldahhide. The suit alleges these emissions are tied to increases in asthma, respiratory diseases, and certain cancers, and alleges violations of the Clean Air Act, specifically in communities with significant black populations. Part of how those gas turbines were initially installed brings up a fascinating regulatory detail. At first, the temporary turbines avoided strict environmental permits because local health departments only regulated gas burning generators if they remained in the exact same location for more than 364 days. However, in January 2026, the EPA actually revised its new source performance standard, ruling that large methane gas turbines now absolutely require permits, even if they are only set up for temporary operations. Let's move to and see how this builds into section 4, legal battles and future. As you can probably imagine, the sheer strategic value of this AI infrastructure has sparked some highstakes clashes. In June 2026, the executive branch, specifically the Trump administration's Department of Justice, moved to intervene and actually dismiss the NAACP's lawsuit. The administration presented the defense that the lawsuit threatens American national, economic, and energy security. They basically argued that shutting off the power supply for this artificial intelligence innovation would interfere with support for Department of War military operations. The legal argument presented here by the executive branch, signed by Stanley Woodward, Jr., is quite specific. It claims that the executive branch has the right to unilaterally dismiss a case brought under the Clean Air Act by an outside party if the government objects to it. They cited executive orders 14156 and 14179 positioning the protection of AI innovation as a matter of critical national importance that supersedes the citizen enforcement actions of the environmental lawsuit. But despite the ongoing legal battles, XAI has been taking physical steps to mitigate their environmental footprint and decouple from that strained local grid. Back in November 2024, they successfully upgraded their grid connection to 150 megawatts, which allowed them to remove some of those temporary gas turbines. They've also deployed 168 Tesla megapac battery storage units at the site to help manage the power draw. And as we discussed a bit earlier, they are pushing forward with building that $80 million dedicated wastewater recycling facility, which leaves us with this final thought. We are literally witnessing one of the most rapid and astonishing technological buildouts in human history. But as we race to solve real world AI, how do we actually balance this unprecedented technological triumph against its massive permanent physical footprint? It's a huge question and it's one that engineers, communities, and lawmakers are going to be grappling with for years to come. Thank you so much for joining me for this explainer. I really hope it helped you visualize the true scale of artificial intelligence. Keep learning and I'll see you next time.