IBM Cloud Set to Launch Intel’s Gaudi 3 AI Chips in 2024

Intel has secured its first cloud customer for the Gaudi 3 AI accelerator chip: IBM Cloud.

On Thursday, IBM and Intel announced that IBM Cloud will start offering the Gaudi 3 to customers early next year. This advanced chip will be available for both hybrid and on-premise solutions, and IBM plans to integrate Gaudi 3 support into its Watsonx AI and data platform.

“Unlocking AI's full potential demands an open, collaborative ecosystem providing customers with a variety of accessible solutions,” said Justin Hotard, general manager of Intel’s data center and AI division. “By integrating Gaudi 3 AI accelerators with Xeon CPUs in IBM Cloud, we are enhancing AI capabilities while addressing the need for cost-effective, secure, and innovative AI computing solutions.”

Unveiled in December 2023, Gaudi 3 is Intel's response to competing AI chips from Nvidia and AMD. It marks the conclusion of the Gaudi series, a portfolio Intel developed through its $2 billion acquisition of Habana Labs in 2019. Earlier this year, Intel showcased reference designs for Gaudi 3 that companies like Lenovo, Dell, HPE, and Supermicro can implement in their servers. These designs feature a new Ethernet connectivity intended to rival Nvidia's InfiniBand technology and are optimized to work with Intel’s Xeon 6 processor series.

However, Gaudi 3 enters a competitive landscape where Nvidia has a strong foothold. In April, Intel projected $500 million in revenue from Gaudi 3 for 2024—a stark contrast to AMD’s anticipated $4.5 billion from its Instinct MI300-series GPUs and Nvidia's expected $40 billion from its data center business. Although Gaudi 3 demonstrates commendable performance for its cost according to early benchmarks, attracting customers with established ties to Nvidia remains a significant challenge.

In July, Intel's CTO Greg Lavender expressed optimism about potentially claiming second place in the AI chip market behind Nvidia. Yet, just a month later, Intel announced a $1.6 billion loss for Q2 and plans to cut 15,000 jobs to save $10 billion by 2025.

Compounding Intel's challenges, Nvidia is set to increase production of its next-gen AI chip, Blackwell, in Q4 following a short manufacturing delay. Blackwell is projected to offer performance up to four times greater than the H100, against which Gaudi 3 has been favorably compared. Intel has refrained from making direct comparisons with Blackwell until the chips become publicly available.