Samsung’s recent partnership with AMD marks a significant move in its AI infrastructure strategy, with the company purchasing US$20 million of Instinct MI300X GPUs to power its data centers. This decision reflects Samsung’s emphasis on a cost-effective approach, as AMD’s GPUs offer a far more affordable alternative than those from NVIDIA, which has dominated the market.
The price difference between the two companies’ products is substantial. While NVIDIA reportedly charges between US$30,000 and US$40,000 per H100 GPU, AMD’s MI300X GPUs are priced at around US$10,000 each. With Samsung’s $20 million investment, the math indicates the purchase of approximately 2,000 AMD GPUs, a significant acquisition that underscores the cost advantages AMD offers in AI hardware.
Despite the price gap, AMD’s MI300X offers excellent value for money. It features advanced specifications, including 19,456 stream processors, 304 compute units (CUs), and a hefty 192 GB of HBM3 memory, making it highly capable of handling demanding AI workloads.
This blend of cutting-edge performance at a lower price point makes AMD a strong competitor in high-performance computing, particularly for companies like Samsung looking to optimize costs without sacrificing performance.
How will Samsung use AMD’s MI300X GPUs?
Samsung’s shift toward AMD’s Instinct MI300X GPUs for its AI data centers appears to be driven by the rising costs and scarcity of NVIDIA’s H100 GPUs, exacerbated by component shortages. The increased demand for H100s led to a challenging situation for Samsung employees, who reportedly had to join waiting lists to access the GPUs, with only 16 units being made available to the team at one point. This scarcity of critical AI hardware made it difficult for Samsung to scale its AI initiatives.
In response, Samsung seems to have found a solution by partnering with AMD, with whom it already agreed to supply HBM3 memory. This partnership extends into AI hardware, with AMD committing to a steady supply of MI300X GPUs in the coming years. The Instinct MI300X offers competitive performance and is significantly more affordable than NVIDIA’s H100, allowing Samsung to sidestep the bottleneck caused by NVIDIA’s supply issues.
While Samsung’s plans for these GPUs are not fully disclosed, it’s reasonable to assume that they will be used to expand AI training and development as part of Samsung’s broader market strategy, including the growth of Galaxy AI. However, the transition won’t happen immediately. According to insiders in the semiconductor industry, Samsung has just started receiving and testing the MI300X units.
After validation, the company will gradually integrate these GPUs into its data centers, a process that will likely take some time to implement fully.
This move aligns with Samsung’s desire to ensure long-term scalability in AI technology, using AMD’s more cost-effective hardware as a key component in its future data infrastructure.
