AMD (AMD, Financial) announced its plan to begin mass production of the new MI325X AI chip in the fourth quarter, aiming to enhance its competitiveness in a market dominated by NVIDIA (NVDA). This announcement was made during an event in San Francisco, as major tech companies like Microsoft (MSFT) and Meta (META) witness AI processor demand exceeding supply.
The MI325X chip utilizes the same architecture as last year's MI300X and features innovative memory technology that boosts AI computing speeds. It is designed to compete with NVIDIA's Blackwell architecture and includes up to 256GB of HBM3E memory, an increase of 64GB from the MI300X, with bandwidth rising from 5.3TB/s to 6TB/s. The core structure remains unchanged, featuring 5nm XCD modules, 6nm IOD modules, 3.5D packaging, 153 billion transistors, and 304 computing units. Notably, the MI325X's power consumption is 1000W, up 750W from the MI320X, and supports eight chips working in parallel, forming a powerful platform with 2TB memory and 48TB/s bandwidth.
Despite the launch, AMD's stock price experienced a slight decline during the event. The company also unveiled new network chips to accelerate data transfer within data centers and introduced a new generation server CPU, codenamed "Turin," dedicated to supporting GPUs and speeding up AI processing. This flagship chip, priced at $14,813, contains nearly 200 processing cores and utilizes the Zen 5 architecture to deliver up to 37% speed enhancement for advanced AI data computations.
Although AMD released new products, they are not expected to significantly impact NVIDIA's data center revenue in the short term. In July, AMD revised its AI chip revenue forecast from $4 billion to $4.5 billion, benefiting from the generative AI boom and strong demand for the MI300X chip. Analysts expect AMD's data center revenue to reach $12.83 billion this year, whereas NVIDIA's forecast stands at $110.36 billion.
As of the close of the event, AMD's stock price decreased by 4%, closing at $164.18.
