South Korean AI chip startup Rebellions is making strides in the AI infrastructure market with its new rack-scale compute platform. This initiative follows the company’s successful establishment in its domestic market and aims to facilitate global expansion without requiring enterprises to adopt liquid cooling or ultra-power dense racks.
Founded in late 2020, Rebellions has focused on developing AI accelerators that have found applications in various sectors within South Korea, including telecommunications, service providers, and enterprise users. Marshall Choy, the company’s chief business officer, stated, “We built up use cases around everything from call centers and customer service to CCTV surveillance for the national highway system.” This experience positions Rebellions to replicate its successes in international markets.
Rebel100 AI Accelerators
At the core of Rebellions’ offering is the Rebel100 AI accelerator, which has been compared to Nvidia’s H200 accelerators. The Rebel100 is capable of achieving a petaFLOP of dense 16-bit floating point math, or double that at FP8. Unlike Nvidia’s monolithic compute die, the Rebel100 employs a chiplet architecture featuring four compute dies manufactured by Samsung. This design is supported by four HBM3e stacks, providing a total of 144 GB of capacity and an impressive 4.8 TB/s of aggregate bandwidth.
Rebellions has opted for a PCIe card form factor with a thermal design power (TDP) of 600 watts, allowing for eight cards to be integrated into a single air-cooled node. This design choice emphasizes compatibility with existing enterprise datacenters, contrasting with Nvidia’s latest liquid-cooled solutions.
RebelRack and RebelPod Systems
The RebelRack system will consist of four nodes, each supporting eight Rebel100 accelerators, resulting in a total of 32 accelerators capable of delivering 64 petaFLOPS of FP8 compute, along with 4.6 TB of HBM3e and 153.6 TB/s of memory bandwidth. For larger deployments, Rebellions is also developing the RebelPod, which can scale from eight to 128 nodes interconnected via 800 Gbps Ethernet.
Software and Future Plans
Rebellions’ software stack is built on open-source frameworks such as vLLM, PyTorch, and Triton, facilitating ease of use for developers familiar with these technologies. Choy emphasized the company’s commitment to open-source solutions, stating, “Everything’s open source, from vLLM compiler all the way up to the very highest level of stack, Red Hat, OpenShift, and everything in between.”
Recently, Rebellions secured $400 million in a pre-IPO funding round, which will support its expansion and the development of more advanced AI systems. The company is expected to file for an IPO within the year, marking a significant milestone in its growth trajectory.
This article was produced by NeonPulse.today using human and AI-assisted editorial processes, based on publicly available information. Content may be edited for clarity and style.
