AMD has unveiled its new AI chip, Instinct MI325X, aiming to challenge NVIDIA's dominance in the $500 billion data center GPU market. AMD plans to release new chips every year.

The Instinct MI325X is designed to compete with NVIDIA’s data center GPUs and is expected to enter production by the end of 2024. Starting from Q1 2025, companies like Dell, HP, and Lenovo are expected to integrate the chip into their systems.

MI325X production is slated to begin later this year

This move could disrupt NVIDIA’s pricing strategy. NVIDIA has been enjoying a 75% gross margin due to high demand for its GPUs over the past year. AMD claims that its 256 GB Instinct MI325X GPU outperforms NVIDIA's 141 GB H200 processor in AI inference workloads.

Powered by CDNA 3 architecture, the MI325X offers 256 GB of HBM3E memory with 6.0 TB/s bandwidth, providing 1.8x more capacity and 1.3x more bandwidth than H200. It also delivers 1.3x higher peak compute performance in FP16and FP8. These advantages translate into up to 1.3x better inference performance on Mistral 7B, 1.2x on Llama 3.1 70B, and 1.4x on Mixtral 8x7B models.

The release of Instinct MI325X marks AMD's push to rival NVIDIA in the data center GPU market. AMD aims to capture a significant portion of this market, which is expected to be worth $500 billion by 2028.

AMD CEO Lisa Su said at the product announcement, "The demand for AI continues to grow and exceeds expectations." While AMD shares have gained only 20% in 2024, NVIDIA has seen a remarkable 175% rise.

With the launch of MI325X, AMD is accelerating its product launch cycle to release new chips annually, increasing competition with NVIDIA and capitalizing on the AI chip boom. The new AI chip replaces the MI300X launched late last year. MI350 is expected in 2025, followed by MI400 in 2026.