Elon Musk's Colossus 2 to Deploy 1.2 Million Nvidia AI Chips by Year-End

Elon Musk Accelerates Nvidia AI Chips Deployment at Colossus 2

Elon Musk announced an aggressive timeline to more than double the Nvidia AI chips count at his Colossus 2 data center near Memphis, Tennessee, aiming to reach over 1.2 million processors by late December 2026. The SpaceXAI CEO revealed that the facility currently operates 110,000 Nvidia GB200 chips and 440,000 Nvidia GB300 chips1

1

. Another 220,000 GB300 chips will become fully operational next week, followed by an additional 220,000 in November3

3

. "If we get lucky, yet another 220k GB300 by late December," Musk wrote on X4

4

. This schedule would bring Colossus 2 to 1.21 million GB200 and GB300 processors, including 1.1 million GB300s.

World's Largest AI Supercomputer Expansion Strategy

The Colossus data center represents Musk's ambitious play in the AI arms race, positioning SpaceXAI to compete directly with established players like OpenAI and Anthropic. Combined with Colossus 1, which houses 150,000 H100, 50,000 H200, and 30,000 GB200 GPUs, SpaceXAI's total GPU count would reach 1.44 million units1

1

. This achievement is remarkable considering SpaceXAI is only three years old, while Anthropic and OpenAI are six and ten years old, respectively2

2

. Musk believes that if the company's "second derivative remains strong," SpaceXAI will reach pole position in about six months2

2

.

Blackwell GPUs Power Next-Generation AI Infrastructure Expansion

Colossus 2 focuses primarily on Nvidia's Blackwell GB300 chips, which are advanced variants of the GB200 featuring the same Grace CPU but a newer GPU dubbed Blackwell Ultra2

2

. The GB300 NVL72 platform connects 72 Blackwell Ultra GPUs in a liquid-cooled rack using fifth-generation NVLink, with each rack drawing up to 142 kilowatts and providing 130 terabytes per second of NVLink bandwidth3

3

. Interestingly, Colossus 1, which features a combination of Hopper and Blackwell GPUs, proved inefficient for training the Grok AI model, so Musk rented it out to Anthropic for inference instead1

1

. Colossus 2 exclusively uses Blackwell GPUs, ensuring there won't be any bottlenecks.

Power Supply Challenges Drive 1.2-Gigawatt Power Plant Construction

One of the biggest challenges facing AI compute infrastructure buildouts is power availability, and SpaceXAI addressed this by developing its own energy solution. The company is constructing a 1.2-gigawatt power plant to support the massive data center capacity1

1

. This approach has generated controversy, with SpaceXAI facing a lawsuit for running unpermitted gas turbines. The company promised to remove these temporary turbines over a one-year span as its permanent power plant comes online1

1

. Musk emphasized that "bringing massive compute online rapidly is incredibly difficult," highlighting that large AI data centers require power plants, transformers, liquid-cooling loops, and complex networking3

3

.

Commercial AI Compute Business Emerges from Internal Infrastructure

Beyond training the Grok AI model, Colossus has evolved into a commercial compute business. Reflection AI agreed to pay SpaceXAI $150 million per month for immediate access to GB300 capacity at Colossus 23

3

. Google signed a multiyear agreement covering about 110,000 Nvidia GPUs, with a $920 million monthly price at full scale, while Anthropic gained access to more than 300 megawatts at Colossus 13

3

. This commercial strategy helps offset the enormous capital expenditure required for the buildout while generating substantial revenue streams.

Long-Term Vision Extends to 50 Million GPUs by 2030

Musk's ambitions extend far beyond the current 1.2 million chip target. He stated that SpaceXAI will grow its data center capacity sevenfold by 2027, aiming for 50 million H100-equivalent GPUs by 20301

1

. The expansion won't be limited to ground-based facilities either. SpaceX is planning to launch an Orbital Data Center System with a million satellites, though Nvidia CEO Jensen Huang characterized this as a "dream" for now1

1

. The AI arms race is shaping up as the most expensive corporate battle of the 21st century, with the belief that the first to the finish line will dominate the market, making speed crucial5

5

. Watch for how SpaceXAI's computational capacity translates into competitive AI model performance against OpenAI's GPT-6 and Anthropic's Fable class models in the coming months.