Rebellions to connect Korean AI chips to NVIDIA's network with SK Telecom

NVIDIA, Korean chipmaker Rebellions, and SK Telecom announced plans to connect Rebellions' upcoming R200 chips, called NPUs, to NVIDIA's computing platform using NVLink Fusion, a high-speed connection technology, along with NVIDIA's MGX rack hardware design. The planned setup would divide AI inference work between the two companies' chips: NVIDIA GPUs would handle the initial processing stage, while Rebellions' chips would handle the following stage, aiming to efficiently run extremely large AI models with trillions of parameters. This approach reflects a growing trend where companies build hybrid systems combining different chipmakers' hardware rather than relying on a single vendor for an entire AI pipeline. Rebellions expects its R200 systems to be ready for real-world deployment by 2028, positioning this as a multi-year infrastructure bet tied to South Korea's push for sovereign, locally built AI computing capacity.

Why it matters

Signals growing interest in mixing different chipmakers' hardware to run massive AI models more efficiently.

Source: NVIDIA