AMD has introduced its next-generation Instinct MI400 AI accelerator, positioning it as a major leap in the company’s data-center hardware lineup. The chip is designed for training large language models and, according to AMD, can deliver up to four times the performance of the current MI300X.

The announcement underscores the intensifying race among semiconductor makers to supply the booming artificial intelligence market. As demand rises for faster model training and more efficient compute infrastructure, chip vendors are pushing new architectures aimed at improving speed, scale, and energy use.

AMD’s MI400 will be closely watched by cloud providers, enterprise customers, and AI developers looking for alternatives in a market long dominated by rival offerings. The company is betting that higher performance and continued gains in efficiency will help it win a larger share of AI infrastructure spending.

While the performance claims come from AMD and will need to be tested in real-world deployments, the launch marks another step in the rapid evolution of AI hardware. For businesses building and running large models, the competition among chipmakers could help drive more options and faster innovation.