Tesla FSD Cortex AI supercluster is running at full capacity, confirms Elon Musk

Tesla (TSLA) CEO Elon Musk has confirmed that the Cortex supercluster is running at full capacity as the demand for video data processing and AI output for the Tesla FSD fleet is increasing each day.

A video clip from Giga Texas local observer Joe Tegtmeyer’s aerial coverage of the factory was shared on X. This footage showed the giant cooling fans installed for the Cortex 1 supercluster building exhausting heat from the huge AI data center.

Last year, Musk said that at full capacity, this huge data center will have a combination of around 100,000 Nvidia H100 and H200 GPUs. “This will be ~100k H100/H200 with massive storage for video training of FSD & Optimus,” he wrote on X.

True
— Elon Musk (@elonmusk) June 20, 2025

– Advertisement –

Aerial view of the Tesla Cortex FSD AI Supercluster cooling tower fans at Giga Texas (Cortex 1). Credit: Joe Tegtmeyer / YouTube.

The building seen in the photo above houses 6 large pipes and huge fans to cool down GPU cluster that’s used for the video training of Tesla Full Self-Driving (FSD) & Optimus humanoid robot’s AI Vision. The first supercluster at Giga Texas (located in front of the cooling plant) is referred to as Cortex 1.

The 2nd supercluster building is under construction at Giga Texas, which is known as Cortex 2. According to the latest updates by Joe Tegtmeyer, only the ground structure and roof of the Cortex 2 building are completed so far (watch videos below).

Only one supercluster is not enough to compute all the video training data, as Tesla is building another supercluster (Cortex 2) at Giga Texas.

Tesla’s large fleet of global cars provides video clips to the Tesla data centers in the United States. This video training data, coupled with AI teaches Tesla FSD and Optimus bots how to identify objects, humans, and the rest of the world around them.

Last year, Elon Musk gave a short tour of the inside of the Cortex 1 supercluster as it was nearing completion (video below).

– Advertisement –

Video of the inside of Cortex today, the giant new AI training supercluster being built at Tesla HQ in Austin to solve real-world AI pic.twitter.com/DwJVUWUrb5
— Elon Musk (@elonmusk) August 26, 2024

Tesla’s Cortex 1 supercluster is one of the world’s largest AI data centers. The closest is also Elon Musk-owned xAI’s Colossus supercluster that’s used to train the Gork AI chatbot.

For perspective, the Cortex GPU cluster is almost 10 times larger than Tesla’s Dojo supercluster. However, Dojo has a combination of Nvidia H100s and Tesla AI4/HW4 processors, and it will have the next-gen AI5 processors in the future, while Cortex currently relies solely on Nvidia’s H100 and H200s.

Although the Cortex supercluster is huge, the Tesla AI Vision data center building is just a fraction of the humongous size of the Giga Texas factory. Giga Texas houses manufacturing for the Model Y electric SUV and the Cybertruck. The Cybercab manufacturing is also going to be housed at Giga Texas.

According to Tesla’s official software release notes, the automaker achieved a 5x training compute scaling enabled by Cortex. So, another Cortex supercluster would add the same amount of compute power.

However, these figures came out when Cortex 1 only had ~50,000 GPUs installed (50% capacity). At 100% capacity with ~100k GPUs, Cortex should add 10x training compute. So, 2 clusters would ideally add 20x training compute for FSD and Optimus bot’s vision AI.

Tesla’s AI training compute demand is constantly growing as more and more vehicles get into customer hands. The launch of the Robotaxi service is also going to increase the load. Fortunately, Tesla is already working on the Cortex 2 data center (no completion estimates yet).

Construction progress of the Cortex 2 data center at Tesla Giga Texas as of 25th June 2025. Credit: Joe Tegtmeyer / YouTube.

– Advertisement –

Video: Giga Texas drone coverage from 20th June 2025 with Cortex 1 and Cortex 2 updates.

Water Usage by Cortex Supercluster Cooling System

There has been a debate on how much water the Tesla Cortex supercluster’s cooling system uses. Some rough estimates by online users suggest it’s using as much as 8 million gallons of water per day to cool down the 100k GPUs running 24×7.

However, Joe Tegtmeyer contradicts these claims. The Cortex 1 cooling system mostly uses reclaimed rainwater, and the building has 4 huge water tanks beneath it for water collection and storage for use by the cooling towers.

Most of this water is recycled, which can be seen in the large ponds beside the Cortex cooling plant. Some of the water is lost in the form of vapors (can be seen in videos and pics above). This vaporized water is far from 8 million gallons a day.

Tesla hasn’t provided any official numbers of how much water it uses for the Cortex cooling system.