NVIDIA CEO Jensen Huang met this week with Rebellions co-founder and CEO Sunghyun Park at NVIDIA's Santa Clara headquarters to discuss potential collaboration. According to Bloomberg, talks cover a spectrum of options: technical partnership, equity investment, or outright acquisition. Bundang, South Korea-based Rebellions has raised ~$850 million from SK Hynix, Samsung Ventures, Arm Holdings and direct South Korean government backing, and is currently valued at ~$2.3 billion.
Rebellions, founded in 2020, specializes in neural processing units (NPUs) optimized for data center AI inference—a workload distinct from NVIDIA's core training GPUs. Rebellions recently secured a $400 million pre-IPO round in March 2026 and is preparing for a listing on South Korea's main stock exchange (KOSPI) in late 2026 or early 2027. The company is also planning a US ADR listing after its South Korea debut.
The deal structure echoes NVIDIA's recent playbook: in late 2025, NVIDIA licensed Groq's technology for a non-exclusive fee, hired most of Groq's engineering team, and kept Groq operating as an independent entity. Rebellions would likely follow a similar hybrid arrangement if talks advance. Any acquisition or acquisition-like deal would face scrutiny from the US Department of Justice (NVIDIA's dominance in training chips draws antitrust focus) and from Seoul, which treats advanced semiconductors as strategic national assets.
For architects, this signals NVIDIA's continued focus on securing inference-layer technology and talent. Inference is now the larger-volume workload in operational AI deployments, and NVIDIA is making strategic bets across the stack: licensing model factories (Poolside $6B), building LPU accelerators (Groq 3), and exploring energy-efficient NPU designers (Rebellions). Early-stage talks mean a deal is not certain, but the engagement itself shows how critical inference is to NVIDIA's infrastructure strategy.