Elon Musk’s AI startup xAI launches “Colossus,” the world’s most powerful AI cluster powered by 100,000 Nvidia H100 GPUs

Less than a year after its inception, Elon Musk’s AI venture, xAI, has launched “Colossus,” now the world’s most powerful AI cluster on the planet, powered by 100,000 Nvidia H100 GPUs. The entire system was built in just 122 days, with plans to double its size soon.

Colossus, featuring 100,000 liquid-cooled Nvidia H100 GPUs linked on a single network fabric, is recognized as the world’s most powerful, according to Musk. This rapid development was completed over a holiday weekend in the United States, a remarkable achievement.

xAI is set to expand Colossus to 200,000 GPUs, including 50,000 H200s, in the coming months. In a recent podcast, Elon Musk, the founder of xAI, mentioned that Grok 2 was trained on just about 15,000 GPUs.

World’s Most Powerful AI Training Cluster

In a post on X (formerly Twitter), Elon said that xAI’s Colossus, the world’s most powerful AI training cluster, is now online after being built in just 122 days. The system will double in size to 200,000 GPUs in the coming months.

“This weekend, the @xAI team brought our Colossus 100k H100 training cluster online. From start to finish, it was done in 122 days. Colossus is the most powerful AI training system in the world, and it will soon double in size to 200k (50k H200s) in a few months. Great work by the team, Nvidia, and our many partners/suppliers,” Musk said on X.

This weekend, the @xAI team brought our Colossus 100k H100 training cluster online. From start to finish, it was done in 122 days.

Colossus is the most powerful AI training system in the world. Moreover, it will double in size to 200k (50k H200s) in a few months.

Excellent…

— Elon Musk (@elonmusk) September 2, 2024

xAI’s Grok 2 recently matched OpenAI’s GPT-4 in record time, using only around 15,000 GPUs. With more than six times that number now in use, xAI and future versions of Grok are poised to challenge OpenAI, Google, and others. In July, Musk claimed xAI is training “the world’s most powerful AI by every metric.”

The news of the launch comes just two months after xAI ended discussions with Oracle over a potential $10 billion server deal. The talks aimed to expand an existing arrangement where xAI had been renting Nvidia AI chips from Oracle’s cloud services for xAI’s data center in Memphis, Tennessee, according to multiple sources involved in the discussions.

To fuel its expansion, xAI secured $6 billion in funding in May, boosting its valuation to $18 billion. The round was led by Andreessen Horowitz, with participation from Lightspeed, Sequoia, and Tribe

Launched last year, xAI aims to deliver “maximum truth-seeking AI.” The earlier rollout of Grok for Premium+ subscribers underscored Musk’s goal to shake up the AI industry, a sentiment he reiterated in a podcast with Lex Fridman, where he criticized OpenAI’s shift from its nonprofit, open-source beginnings.