gradient

HPE may have been beaten Supermicro and Dell to win a $1bn AI contract, but it’s not for the Colossus supercomputer

Bloomberg reports the contact is for Elon Musk’s X social media platformxAi’s Colossus uses 100,000 Nvidia H100 GPU and doesn’t seem to be part of the contract Supermicro has had troubles, but is still a major player in AI hardware

The rising demand for computing power to support AI workloads has fueled rapid growth in the market for high-powered servers. It is a lucrative business for firms like Dell, Supermicro, and Hewlett Packard Enterprise, who have all seen increased demand for their high-performance server products in recent years.

A report from Bloomberg claims HPE has secured a massive deal worth more than $1 billion to supply Elon Musk’s X social network with servers optimized for artificial intelligence work.

The report doesn’t specify exactly how the servers will be used, but since it’s for X, there’s a good chance some of the capacity will be for Grok, the social network’s AI chatbot. At the end of 2024, X announced that it was rolling out Grok-2 to all users for free, and it’s fair to conclude that extra capacity will require more server infrastructure.

Dell and Supermicro outbid

Bloomberg says X’s agreement with HPE was reached in late 2024, according to people familiar with the matter.

Musk’s companies, including xAI and Tesla, are major buyers of AI hardware. The Colossus supercomputer, built by xAI in Memphis, was previously declared by Musk to be the “most powerful AI training system in the world.” The billionaire stated it was built “from start to finish” in just 122 days.

Colossus uses 100,000 Nvidia H100 GPUs, but there are plans to massively expand its operations. Supermicro has faced some challenges recently, including the resignation of its auditor and other related matters. However, as MarketWatch reports, it remains a “major player in the AI revolution” and recently established operations in Memphis to support xAI’s goal of building a supercomputer facility with 1 million GPUs.

Supermicro and Dell provided the servers for Colossus originally, and Bloomberg says that while both firms did bid to supply the equipment for this new venture they were ultimately unsuccessful.

Bloomberg’s report notes, “HPE’s liquid-cooling technology may have played a role in the win,” wrote Woo Jin Ho, an analyst at Bloomberg Intelligence. While good for sales, larger deals can be a drag on margins, he added.”

You might also like

Tesla will spend billions on Nvidia and AMD GPUs and CPUsElon Musk shows us what $1 billion worth of AI GPUs looks likexAI cluster is now the most powerful AI training system in the world