Rebellions and FuriosaAI showcase server inference chips for physical AI
The AMW Read
The exhibition incrementally updates two inference-chip suppliers' physical-AI positioning, with hardware architecture central but no robotics deployment metrics demonstrating broader impact.
Named counterparties: FuriosaAI
Rebellions and FuriosaAI showcase server inference chips for physical AI
Rebellions and FuriosaAI displayed inference hardware at Physical AI Tech Day in Seoul on October 14, showing how servers can process inputs from robots. Rebellions exhibited Atom Max and RebelCard, which carries its Rebel100 NPU. According to company materials cited in the report, Rebel100 combines four compute chiplets with 144GB of HBM3E memory; RebelServer houses eight RebelCards. FuriosaAI showed its second-generation Renegade NPU card, equipped with 48GB of HBM3 memory and a PCIe connection. Its developer documentation lists support for language, multimodal, and vision-model inference.
The market significance lies in the infrastructure beneath physical AI: some perception and decision-making workloads can run on connected servers outside the robot. That gives server inference suppliers a potential role in robotics without requiring their chips to sit inside the machine. Both companies also provide compilers and runtime software, making model compatibility and execution support part of the competitive offering alongside chip architecture and memory capacity. The report describes hardware positioning, however, and provides no measured robotics latency, deployment volumes, or comparative performance results.
Builders evaluating these cards should test their intended models across the complete camera-to-network-to-server path, measuring response time and confirming software support. Investors should distinguish product availability and integration partnerships from demonstrated robotics adoption. FuriosaAI's March enterprise-server collaboration with LG Uplus, combining Renegade, EXAONE 4.0, and the ixi Enterprise platform, illustrates a route to packaged deployment; the article does not establish that product as a robotics installation.
