by RebellionsKR
Deliver energy-efficient AI inference at lower power and cost than GPU-based systems.
In development. POC trials with Meta and xAI.
Checked 11 Oct0 updates this week
Rebel100 uses chiplets that combine to form inference systems, allowing flexible configuration for different workload sizes.
The chip integrates with PyTorch, Kubernetes, and Hugging Face, minimizing customer software integration work.
| Spec | Rebel100 |
|---|---|
| Performance per watt claimed | 3.8x GPU baselineR (reported) |
| Capital expenditure advantage | 1.9x lower vs GPU nodesR (reported) |
| Operating expenditure advantage | 2.1x lower vs GPU nodesR (reported) |
R reported by the company
Rebellions
Rebel100 chip, RebelRack, AI inference, Korea
South Korea fabless semiconductor company for AI inference. Founded 2020, merged with Sapeon 2024. Rebel100 NPU with 3.8x performance-per-watt. March 2026 raised $400M at $2.34B valuation.
