Samsung will build an AI Megafactory using 50,000 Nvidia GPUs

After Samsung Electronics chairman Jay Y. Lee met Nvidia CEO Jensen Huang yesterday, the two companies announced a massive partnership. One of the highlights of the partnership is Samsung's plan to use over 50,000 Nvidia GPUs to build massive AI factories in South Korea.

Samsung has just announced that it will create a giant AI factory in partnership with Nvidia. In a shift towards AI-driven manufacturing, it plans to deploy more than 50,000 Nvidia GPUs. These GPUs will be used throughout Samsung's entire existing manufacturing flow. They will also be used in improving the development of smartphones (and other mobile devices), semiconductor chips, and robotics-based products.

The South Korean firm plans to use Nvidia's GPUs in its semiconductor chip design and manufacturing processes. AI will be used to improve the company's design, manufacture, and quality control processes. All these processes will be connected to a single network where AI will continuously analyse, monitor, and optimise production in real time.

Samsung claims it has already achieved more than 20x improvements in computational lithography performance using Nvidia's cuLitho and CUDA-X libraries for its optical proximity correction (OPC) process. The two companies are also working to develop next-generation GPU-accelerated electronic design automation (EDA) tools and design technologies.

The two companies have been working together for more than 25 years. It all started with Samsung's DRAM chips powering Nvidia's earliest graphics cards. Samsung's foundry division also makes Nvidia's GPU chips.

Samsung said it is also working with Nvidia for HBM4 chips that offer incredible speeds and high power efficiency. Those chips will be made using Samsung's sixth-generation 10nm class DRAM and 4nm logic base die to offer up to 11Gbps data transfer speeds, greatly exceeding JEDEC standards of 8Gbps.

As part of this partnership, Samsung will provide GDDR, HBM, and SOCAMM memory chips to build AI servers that use Nvidia's GPUs. The South Korean firm will also use Nvidia's Omniverse platform to develop digital twins manufacturing platforms.

The South Korean technology giant will use Nvidia RTX PRO 6000 Blackwell Server Edition platform to improve its manufacturing automation and humanoid robotics. It is also using the Jetson Thor robotic platform to enhance real-time AI reasoning, safety controls, and task execution in intelligent robots.

Nvidia will collaborate with Samsung to develop AI-RAN development that will be used by South Korean telecom service providers. This technology will use Samsung’s software-defined network and Nvidia's GPUs.