NVIDIA in Preliminary Talks with South Korean AI Chip Startup Rebellions
NVIDIA has held preliminary discussions with South Korean AI chip startup Rebellions regarding potential technical collaboration, investment, or acquisition. Leaders from both companies met at NVIDIA's headquarters in Santa Clara, California, to explore these opportunities. As the demand for AI hardware surges, NVIDIA's interest in Rebellions highlights the growing importance of specialized NPU architectures for AI inference. A potential deal could strengthen NVIDIA's position in the inference market and expand its influence in South Korea's semiconductor ecosystem. Rebellions specializes in NPU-based AI inference accelerators, with its latest Rebel100 chip boasting 2 PFLOPS of FP8 compute power, 523MB of SRAM, and 144GB of HBM3E memory. However, the talks are in the early stages, and there is no guarantee that any formal agreement will be reached.
## BACKGROUND
Rebellions is a prominent South Korean AI chip startup that focuses on energy-efficient hardware for AI inference rather than training. Neural Processing Units (NPUs) are specialized processors designed to accelerate machine learning algorithms, while High Bandwidth Memory (HBM3E) provides the massive memory bandwidth required to handle large AI models.