xAI’s Colossus 2 – First Gigawatt Datacenter In The World, Unique RL Methodology, Capital Raise

Hamartia Antidote

Elite Member
Joined
Nov 17, 2013
Messages
47,723
Reaction score
27,192
Reputation
532.4
Country of Origin
Country of Residence

Much has been written about xAI’s Colossus 1. The Memphis build belongs in the history books: the largest AI training cluster, erected from scratch in 122 days. With roughly 200,000 H100/H200s and ~30,000 GB200 NVL72, it remains, today, the largest fully operational, single-coherent cluster (setting apart Google, master of multi-datacenter-training).

However, Colossus 1’s ~300 MW looks modest next to the Gigawatt-scale clusters under construction by OpenAI, Meta and Anthropic. Their hyperscaler partners are happy to leverage their balance sheet and win the market by throwing dollars at it.

Was xAI’s prowess a one-time wonder? Today we will publicize some data from our industry leading datacenter model over the last year that is accessible to clients. This is the our same proprietary data that called the Oracle deals many months ahead of the announcement.

image-91.png


Short answer: no. xAI is still squarely in the frontier-AI race and is positioned to leapfrog most rivals again on compute. By our estimates, its total datacenter capacity for a single training cluster will surpass Meta Superintelligence and Anthropic by Q3 2025. The datacenter capacity will be ready for the GPUs to be moved in to create the largest single datacenter in the world, yet again. xAI has to raise the capital for those GPUs, but they have the allocations from Nvidia to have it fully training large scale models early next year.

Elon came up with a new genius trick to beat rivals at time-to-market. Colossus 2 will be an even more impressive achievement than xAI’s first cluster. Let’s dig in.

The first half of this report will dig into the Colossus 2 prowess. The second half will discuss Grok models, our mid-to-long term thoughts on xAI, and the unique RL method xAI is using that may lead them to leapfrog OpenAI, Anthropic, and Google.

Colossus 2: from zero to 200MW in six months​

The Colossus 2 project was kicked off on March 7th, 2025, when xAI acquired a 1m sqft warehouse in Memphis, and two adjacent sites totaling 100 acres. By August 22nd, 2025, we count 119 air-cooled chillers on site, i.e. roughly 200MW of cooling capacity. That’s enough to power roughly 110k GB200 NVL72. And an Elon tweet shows some racks were already installed in July.

xAI built in six months what took 15 months for Oracle, Crusoe and OpenAI!

image-54.png
Source:

Colossus 2: in Tennessee, Mississippi… or both?​

Things get more confusing once factoring-in a Greater Memphis Chamber statement in May that no turbines would be sited in Memphis. They’re not lying.

Memphis and Tennessee have been getting a lot of pushback, so xAI’s genius move was to develop a Gigawatt-scale energy hub right across the border in Southaven, Mississippi. In mid-2025, the company acquired a former Duke Energy power plant in Southaven. Shortly after, Mississippi regulators granted xAI temporary approval to run gas turbines there for up to 12 months without a permit!

image-173.png
Source: SemiAnalysis Datacenter Industry Model
To transport and manage power generated by the Mississippi power plant, xAI is building infrastructure near Colossus 2. We show below the first deployment of Tesla Megapacks, as well as the Medium Voltage power lines connecting the two sites.

image-55.png
Source: SemiAnalysis Datacenter Industry Model

From 200MW to 1.1GW and Solaris Energy Infrastructure partnership​

In Southaven, MS, xAI is moving at the pace of light. The discontinued power plant now sees seven 35MW turbines in operations.

image-174.png
Source:

To deploy faster than peers, xAI relies on rental turbine companies. NYSE-listed Solaris Energy Infrastructure owns a fleet of 600MW of gas turbines, of which ~400MW currently serve xAI. Musk’s firm weighs 67% of SEI’s 1700MW orderbook, i.e. 1,140MW. There are ~240MW on the Memphis Colossus 1 site, while the remaining 900MW will be owned by a Joint Venture owned at 50.1% by Solaris and 49.9% by xAI.

Source: Solaris Energy Infrastructure
As shown below, ~460MW are currently installed and in operations/under construction.

image-52.png
Source: SemiAnalysis Datacenter Industry Model
The newly formed JV already spent $112 million in capital expenditures in Q2 2025. After a slow Q3, spending will pick up again in Q4 2025 and Q1 2026. Solaris expects to have over 1.1GW of fully operating turbines for xAI by Q2 2027. There remains ~425MW available for contracting, and we think xAI will likely pull the trigger to get to over 1.5GW of total gross power. Solaris also appears to temporarily lease power generation capacity from third parties to deliver faster:
 
PHEW!!! 200,000 H100/H200s and ~30,000 GB200 NVL72 can probably alone do most of the entire world's computing calculations. This is massive!
 

XAI Using $18 Billion to Get 300,000 More Nvidia B200 Chips​


On October 5, 2025, the Wall Street Journal reports that xAI is committing over $18 billion to acquire approximately 300,000 additional NVIDIA GPUs, bringing the total for Colossus 2 to around 550,000 chips. This aligns with broader announcements from Musk on X, where he has repeatedly described Colossus 2 as the world’s first gigawatt-scale AI training cluster, with ambitions to scale toward 1 million chips in the near term.
 
smaller turbines are easier to manufacture and install....they literally come in containerized form factors.

but i guess speed is of the essence here.........which makes me wonder how bad of a bubble is it? 2008 mortgage crisis type or the 2000 dot-com types?
 

Tesla Adds Megapacks for Its Next-Generation Supercomputer (Cortex 2)​


With AI continuing to use more of the world’s energy, major AI training deployments need something that no major energy provider can supply - true stability. That stability comes from stationary battery storage - something that Tesla Energy has been deploying in droves.

Now, Tesla Energy has achieved two back-to-back deployment milestones: powering the world’s largest supercomputer in Memphis for xAI, and powering the brains behind the next generation of FSD at its own Giga Texas.

Colossus 2: $585 Million of Batteries​

The sheer scale of xAI’s Colossus 2 supercomputer - already the most powerful in the world - now has an equally immense battery backup to match. xAI and Tesla Energy have recently completed the installation of approximately 600 Megapacks at the site in Memphis, Tennessee.




That represents roughly 2.3GWh of energy storage capacity. For context, that is enough energy to power a small city - or back up the massive grid load required by over 100,000 NVIDIA H100 and H200 GPUs present on site.

Based on current Tesla Energy pricing, this represents approximately $585 million, making it one of the largest battery energy storage projects on the planet. Unlike a typical grid utility project, these batteries aren’t here for grid arbitrage.

Instead, they’re serving as a critical buffer for Colossus and Colossus 2, ensuring that the massive power spikes from AI training runs don’t destabilize the local grid or cause brownouts during peak compute loads.

Cortex 2 in Austin​

Meanwhile, in Austin, the hardware to train the next generation of FSD - both Unsupervised and for Optimus - is arriving on site.

Drone footage captured by Giga Texas observer Joe Tegtmeyer confirms that Tesla has begun taking delivery of Megapacks for its new Cortex 2 Supercomputer. The new Cortex 2 cluster is expected to be a 500MW supercomputer cluster, focused primarily on training for Optimus.

The arrival of Megapacks suggests that Tesla is getting closer to powering up the facility for the first time. Just as in Memphis, these batteries will stabilize the massive 500MW energy load required to train end-to-end neural networks that will allow Optimus to navigate factories, perform complex tasks, and more.

Overcoming Grid Limitations​

These two projects highlight a bottleneck in the AI race: Power. As xAI and Tesla race to build these massive training clusters, the limiting factor is no longer just obtaining enough GPU allocations from NVIDIA. Instead, it’s finding the sheer amount of stable electricity to turn them on.

By vertically integrating the battery storage with the compute, Elon Musk is effectively building his own utility infrastructure to bypass grid limitations.
 
Last edited:
cant wait for the ai bubble to burst :love::love:

Not going to happen. It's already indispensable for productivity and convenience - especially when in voice conversation mode for teaching to make things more interesting for students. It's a Star Trek holodeck.

Examples:


Rome, Italy 100 CE Walkthrough (AI Reconstruction)​



The Great Pyramids of Egypt, 2589 BCE (AI Reconstruction)


Babylon 626 BC Walkthrough (AI Reconstruction)​






xAl's Mind Blowing Grok 4 Demo​


Unfortunately its going to significantly widen the gap between developed and developing countries.
 
Last edited:
Not going to happen. It's already indispensable for productivity and convenience - especially when in voice conversation mode for teaching to make things more interesting for students.

Example:


Unfortunately its going to significantly widen the gap between developed and developing countries.

Gpt Literally will run out of money this year , a ceo crying and telling people to use it more

It will happen
 
Gpt Literally will run out of money this year , a ceo crying and telling people to use it more

It will happen

:rolleyes: Wrong thread. GPT is not what this thread is about. It's about xAI Grok

try this PDF one instead.
"OpenAI launches GPT-5, the worlds most advanced AI model"


AI is getting entrenched quickly. This example shows two AIs interacting with each other.

Conversational xAI Grok AI telling the Tesla driving AI where to navigate the car and things it should do during the trip..all the while having a conversation with the car passenger who has his hands and feet off the pedals and steering wheel.


Here's a conversation with xAI Grok while driving a 2026 Tesla Model S​



Grok seeing what you see and having a conversation about it:

Use video mode: turn on camera in the Grok app and ask Grok to explain what you're seeing.


You can just point your camera at anything and @Grok will tell what it is


GROK CAN SEE WHAT YOU SEE—LITERALLY​


This is just examples of how the tech gap is going to widen very very very quickly.
 
Last edited:
cant wait for the ai bubble to burst :love::love:
For once we agree. AI hype is now years old and questions are being asked about how useful is it really ? The chatbots continue to be useless at handling real world issues and what does AI do that is indispensable ?
 
Wrong thread. GPT is not what this thread is about. It's about xAI Grok
omg xAI is even worse lmao wont get into it u probably know why . ai business Model makes no sense only one that is making money is Nvidia , rest are just draining money .
 
dude the fucking hype , i am so tired of labeling any basic tree /neural algorithm as ai and putting ai into everything .
got so annoying i switched from windows to linux
he chatbots continue to be useless at handling real world issues
i wouldn't say that , they are useful if u know how to use them . just the hype and constant shoving ai into places it doesn't belong is annoying . it going to be like the dot com bubble it will burst and only useful stuff will remain . u live in America prepare for the bubble to burst man .because it will eventually .


 
For once we agree. AI hype is now years old and questions are being asked about how useful is it really ? The chatbots continue to be useless at handling real world issues and what does AI do that is indispensable ?

You are so shortsighted...soon history books will be interactive environments that teachers and pupils will walk in and explore...instead of boring book paragraphs nobody wants to read.


Ancient Mesopotamia - An AI History Short​

 
Last edited:
You are so shortsighted...soon history books will be interactive environments that teachers and pupils will walk in and explore...instead of boring book paragraphs nobody wants to read.


Ancient Mesopotamia - An AI History Short​


This means they will be brainwashed easier and get dumber quicker. Losing the ability to read and imagine will have very deleterious effects on people, far more than reality TV ever did.
 

Users who are viewing this thread

Pakistan Defence Latest

Back
Top