AMD Projects AI Will be at Core of Data Center Boom, Advances GPU Ambitions with Latest Chip

The debut of AMD’s newest chip is their latest attempt to shake up the heavily-concentrated GPU market, now one of the most critical sectors for chipmakers to harness in the era of AI. The MI300X is directly targeted toward the development of large language models like that of the popular ChatGPT and could be a key competitor to Nvidia’s H100 GPU. Though AMD will begin shipping the chip toward the end of the year, the company has yet to announce a price point or any major buyers. Amazon has thrown its own hat in as a potential customer for the new chip, but it remains to be seen if anything official will be announced.

Related ETF & Stocks: iShares Semiconductor ETF (SOXX), Advanced Micro Devices, Inc. (AMD), NVIDIA Corporation (NVDA)

As part of a keynote address at CES 2023, AMD CEO Lisa Su announced that its most-advanced chip for artificial intelligence (AI), the MI300X, will start shipping to some customers later this year. AMD is banking on MI300 family of chips to help the firm compete with Nvidia’s suite of Graphics Processing Units (GPUs), which have become a critical commodity for machine learning capabilities. The MI300X is formed by combining multiple GPU chiplets.

With DRAM memory boosted from 128 gigabytes in the MI300A to 192 gigabytes, the MI300X will be even more useful for the deployment of large language models (LLM) – neural networks like ChatGPT with billions of parameters that are trained on large quantities of unlabeled text data. AMD claims that an LLM like Falcon 40, which contains 40 billion parameters, can now be run on a single, MI300X accelerator5.

Back in March, MRP noted that many of the initial generative AI-based implementations from Google, Microsoft, and other firms appeared to be clunky and error-prone in practice. Therefore, we felt that semiconductors, which form the backbone of large language models and other…