Colossus and Colossus 2 - AI Computer System by Elon Musk
- Core Insights Advisory Services
- Aug 20
- 6 min read
Source: Various and Confidential
Date: August 20, 2025
Elon Musk's xAI has developed a supercomputer named Colossus, which is currently believed to be the world's largest AI training platform. Constructed in Memphis, Tennessee, the initial phase of Colossus was built in just 122 days and originally deployed with 100,000 Nvidia H100 GPUs. The system was brought online in mid-2024 and is primarily used to train xAI’s chatbot, Grok, as well as support AI development for the social media platform X and other Musk ventures like SpaceX.
Colossus features a highly advanced infrastructure, with servers provided by Supermicro and support from Dell Technologies.
Each GPU is equipped with a dedicated 400GbE network interface, and the entire cluster operates on Ethernet rather than InfiniBand, enabling high-bandwidth connectivity across the system. The setup includes liquid-cooled racks and redundant power supplies, with over 1,500 GPU racks making up the cluster.
To manage its immense power demands—estimated at over 150 megawatts for full operation—xAI deployed 14 mobile natural gas generators, each providing 2.5 MW, alongside grid power and Tesla Megapack batteries used as energy buffers. This hybrid power solution was necessary due to insufficient grid capacity during initial deployment.
In a rapid expansion, xAI announced that Colossus had doubled its capacity to 200,000 GPUs, incorporating both H100 and next-generation H200 chips, with plans to scale up to one million GPUs in the future.
This growth is supported by strategic partnerships with Nvidia, Dell, and Supermicro, and reportedly backed by $6 billion in funding.
Nvidia CEO Jensen Huang praised the project as an unprecedented engineering feat, highlighting Elon Musk’s unique ability to orchestrate large-scale technological deployments.
However, the expansion has raised environmental and regulatory concerns in Memphis, particularly regarding emissions from the gas-powered generators.
THERE IS A GLOBAL RACE GOING ON IN COMPUTING TO BE THE COMPANY AND COUNTRY WITH THE MOST, BIGGEST, AND BEST COMPUTING EQUIPMENT TO SUPPORT BOTH CONVENTIONAL COMPUTING AS WELL AS “AI” COMPUTING.
IT IS SIGNIFICANT THAT SEVERAL COMPANIES JOINED TOGETHER WITH THE USA GOVERNMENT TO ATTEMPT TO RACE AHEAD OF THE GIANT “AI” COMPUTING SYSTEM “COLOSSUS” DESIGNED AND BUILT BY ELON MUSK AND HIS xAI COMPANY. AT THE SAME TIME ELON MUSK IS BUILDING A SYSTEM CALLED “COLOSSUS 2” WHICH WILL BE 5 TIMES BIGGER THAN THE FIRST SYSTEM.
THESE PHOTOS SHOW THE SIZE OF THE MUSK “COLOSSUS” “AI” SUPER COMPUTER WHICH COVERS AN AREA OF 785,000 SQUARE FEET.
THE “STARGATE” DATA CENTER PROJECT PLANS CALL FOR FOUR 10 DATA CENTER BUILDINGS ON THE FIRST CAMPUS WITH EACH BUILDING TO BE 500,000 SQUARE FEET MAKING THE TOTAL FOR THE FIRST SITE TO BE 5 MILLION SQUARE FEET. EACH OF THE 19 ADDITIONAL SITES TO BE 5 MILLION SQUARE FEET EACH MAKING A TOTAL OF 100 MILLION SQUARE FEET OF DATA CENTERS DEDICATED TO “AI” EFFORTS. EACH CAMPUS OF 10 DATA CENTER BUILDINGS ARE EXPECTED TO USA 5 BILLION WATTS OF POWER PER CAMPUS WHICH IS ENOUGH TO POWER 5 MILLION HOMES WITH ELECTRIC POWER. THE TOTAL OF 20 GROUPS OF DATA CENTERS IN THE “STARGATE” PROJECT WILL USE 100 BILLION WATTS OF POWER WHICH IS ENOUGH TO POWER 100 MILLION HOMES WITH ELECTRICITY.
THE BOTTOM LINE IS THAT THE MASSIVE SCRAMBLE BY SEVERAL “AI” COMPANIES EVEN BEYOND THE ELON MUSK EFFORTS AND THE JOINT PRIVATE/GOVERNMENT “STARGATE” EFFORT MEANS THAT THE “AI” SYSTEMS ARE EATING UP RESOURCES SUCH AS POWER AS WELL AS WATER FOR COOLING THE DATA CENTERS.
THIS IS A DRAMATIC SITUATION BECAUSE THE USA POWER GRID IS ALREADY SUFFERING FROM NEGLECT AND ABSENCE OF INTENSE LONG TERM INVESTMENT IN PROVIDING POWER TO SUSTAIN THE ECONOMIC GROWTH OF THE USA. AT THE CURRENT RATE OF EXPANSION, THE PRIMARY LIMITING FACTOR IN ALL THE “AI” PROJECTS IN THE USA IS POWER.
AS A NOTE: CHINA IS DOUBLING THEIR AVAILABLE ELECTRIC POWER EVERY 6 MONTHS WHICH HAS CHINA WELL AHEAD OF THE USA EVEN IF THE CHINESE COMPUTING CHIPS USE MORE POWER THAN THE USA CHIPS, CHINA HAS MUCH MORE AVAILABLE POWER TO DEDICATE TO THE “CHINA AI” EFFORT.
IMPORTANT AI NEWS WHICH AFFECTS EVERYONE
New memristor design could be a game-changer for AI and big data. Advanced technologies such as neural networks have found extensive application in image recognition, big data processing, financial analysis, and many other fields. However, training them demands significant computational resources and energy consumption, posing challenges for their widespread use and further development.
The problem comes down to bottlenecks imposed by traditional computing systems built using transistors, which keep their memory and processing units separate, requiring power-intensive and performance-degrading data transfers between them. Perhaps an even greater disadvantage is that their memory requires constant power to store information, which further increases energy consumption.
To solve this, researchers propose an alternative: the memristor. “Memristors, also known as memory resistors, are switches that can ‘remember’ their previous electric state, even after power is switched off,” said Desmond Loke, professor at the Singapore University of Technology and Design, in an email.
“Memristors can be used to create devices that store data because they have the ability to drastically reduce the time and energy required for data transmission between memory and processors in traditional microchips,” he continued. “They might be perfect for constructing neural networks, artificial intelligence systems for medical scan processing, and enabling driverless vehicles.”
In a study published in Advanced Physics Research, Loke and his colleagues investigated neural network training using new memristor design that stores information in a unique material made up of germanium, tellurium, and antimony.
This substance exists in an amorphous phase, but when exposed to an electric current and a change in temperature, it exhibits ordered crystalline regions. Depending on the properties of the pulse, such as its intensity and signal shape, the number and size of these regions changes, consequently affecting the material’s electrical and optical properties, which are preserved after the current is turned off. In this way, information can be recorded in the memristor and read from it by repeatedly applying an electric current.
In total, the authors could create fifteen, clearly distinguishable stages of crystallization in the memristor, each corresponding to specific information necessary for training the neural network. “We used [the] memristors to simulate a neural network,” Loke explained. “In trials, the neural network recognized handwritten numerals with an accuracy of over 96%.” This accuracy is an improvement on other memristor designs by more than 5%, say the team, indicating a promising trajectory for this type of memristor computing technology. But there are still significant hurdles to overcome before this technology can be implemented into real-world computing systems.
“System integration and scale pose considerable challenges,” Loke explained.
Nevertheless, the researchers remain hopeful, envisioning a future where memristors might play a pivotal role in training large neural networks. They anticipate smaller, more potent, and dramatically more energy-efficient units compared to conventional computers if the technology can be developed further.
“Currently, this work is just a proof of concept,” Loke concluded. “The next step is to develop a fully integrated circuit and a big neural network.”
In Summary
A. THE COMPUTING SPEED OF THE VARIOUS “AI” PROJECTS IS AS FOLLOWS:
1. ELON MUSK SYSTEMS:
a. COLOSSUS 1: In early December 2024, Ted Townsend detailed how the power of Colossus doubled in its processing capability. When it first went online[when?], "it was using 100,000 Nvidia H100 processing chips."[30] Still, this initial launch demonstrated Colossus to be the largest supercomputer globally. As of June 2025, the supercomputer got 150,000 H100 GPUs, 50,000 H200 GPUs, 30,000 GB200 GPUs. IT USES 150 MILLION WATTS OF POWER AND MILLIONS OF GALLONS OF WATER DAILY FOR COOLING
b. COLOSSUS 2: JUST DOWN THE STREET WILL HAVE 1 MILLION GB200 GPU’S AND USE 1 BILLION WATTS OF POWER. AND MULTIPLE MILLIONS OF GALLONS OF WATER PER DAY FOR COOLING!
B. PROJECT “STARGATE” 20 CAMPUSES OF 5 DATA CENTERS ON EACH CAMPUS
1.EACH CAMPUS USING THE SAME AMOUNT OF ELECTRICITY AS 5 MILLION HOMES!!!
2. EACH CAMPUS USING ENOUGH WATER FOR 5 MILLION HOMES AND THE FOR WATER NEEDED TO FEED THE PEOPLE IN THOSE HOMES.
3. BOTTOM LINE FOR THE ENTIRE “STARGATE” PROJECT ALONE:
a. THE POWER USED COULD POWER 100 MILLION HOMES
b. THE WATER USED FOR COOLING THE DATA CENTERS IS ENOUGH TO SUPPLY FARMS WITH WATER TO FEED THE 100 MILLION HOMES!!!
C. THE OTHER NON “STARGATE” PRIVATELY OWNED PROJECTS AND DEPARTMENT OF DEFENSE PROJECTS COMBINED WILL USE THESE RESOURCES:
a. THE POWER USED COULD POWER 200 MILLION HOMES
b. THE WATER USED FOR COOLING THE DATA CENTERS IS ENOUGH TO SUPPLY FARMS WITH WATER TO FEED THE 200 MILLION HOMES!!!
D. THIS MEANS THAT FOR FUELING “AI” PROJECTS ONLY, THE RESOURCES USED ARE EQUAL TO 300 MILLION HOMES OF POWER AND THE WATER FOR THOSE 300 MILLION HOMES TO HAVE FOOD!
E. THIS LEVEL OF CONTROL OVER THE PEOPLE OF THE USA MEANS THAT THE PEOPLE HAVE NOT ONLY SURRENDERED THEIR WATER, FOOD SUPPLY, AND THE POWER THEY NEED IN THEIR HOMES EACH DAY!! WHY?
1. ALREADY THE USA POWER GRID HAS PROBLEMS POWERING THE HOMES, STORES, AND FACTORIES IN THE COUNTRY.
2. IN ADDITION THERE IS NOT ENOUGH WATER FOR THE EXISTING FARMS AND HOME, AND THE “AI” COMPUTER PROJECTS ARE DEMANDING HUNDREDS OF MILLIONS OF GALLONS OF WATER PER DAY!