Samsung is creating an AI megafactory together with NVIDIA: the company will connect tens of thousands of video chips and software for «digital twins» into a production network to accelerate the design, testing, and launch of new semiconductors and devices.
Samsung is creating an AI megafactory together with NVIDIA: the company will connect tens of thousands of video chips and software for «digital twins» into a production network to accelerate the design, testing, and launch of new semiconductors and devices.
According to a Samsung press release, the company plans to connect more than 50,000 NVIDIA graphics chips and the Omniverse platform to enable real-time chip design, process tuning, equipment management, and quality control in a single loop. The «digital twins» of factories will allow changes to be driven through a virtual shop floor and predict failures before something goes wrong on the real production line. Details are in the Samsung Newsroom.
NVIDIA libraries for computational lithography will be used in production. This is the stage where it is necessary to «draw» future circuits on wafers with great precision.
The transition to GPUs, according to the company, speeds up these calculations many times, so the preparation of new technical processes becomes shorter and the error is smaller. In parallel, Samsung and partners are preparing a new generation of HBM4 memory, designed for high speeds and energy efficiency in AI systems.
The AI megafactory is not a single building, but a holistic approach: shared data, unified tools, and automated AI guidance for engineers and operators. The company plans to expand this model to its key sites, including in the US (Taylor), to scale production faster and transfer settings between plants more quickly. In robotics, Samsung will deploy NVIDIA platforms to train and control «smart» robots that can make decisions closer to the workplace.
Telecoms has also become an important area. Together with NVIDIA, mobile operators and academic institutions, Samsung is developing AI-RAN: when AI calculations are performed directly on the network, and not only in the «cloud». This is useful for drones, warehouse robots and industrial automation, where millisecond delays can affect safety and performance.
Previously, dev.ua wrote about how Samsung and OpenAI agreed on broad cooperation: the tech giants will develop infrastructure for artificial intelligence, in particular, they will work on projects for floating data centers on water and separate floating power plants to power them.
