AI Chip Developer Rebellions AI Secures Backing from Samsung Ahead of IPO

South Korean AI Chip Developer, Rebellions AI, led by Sunghyun Park(CEO and Co-Founder), Jinwook Oh(Co-Founder and CTO), Hyo-Eun Kim(Co-Founder and CPO), Sungkyue Shin(CFO), secures backing from Samsung as part of a funding round that could be up to $200 million, as reported in CNBC.

Started in 2020, Rebellions has since raised $220 million, valuing the company at $1 billion. The current funding round will surpass the company's valuation of $1 billion.

Rebellions' Atom Series, currently in production, powers full-stack AI with 4 models, Atom Max(boosted performance for Large Scale Inference, featuring direct card-to-card communication), Atom(Cost-Efficient, Powerful AI Acceleration for Small-Sized Data Centers and Enterprise Applications, featuring low latency, Optimized for SLMs, small-and-middle sized enterprise applications), Atom Lite(Low-Power, Yet Highly Powerful AI Inference at the Edge, featuring Low Power Consumption, Compact Design, Optimized for use in workstations, edge environments, or laboratory setups), and Atom SoC(Optimized for Powerful, Low-Latency Al Inference).

Rebellions' second-generation chip, Rebel Quad, is a set of 4 chips, HBM3E Chiplet-Based Scalable Al Accelerator for Hyperscale Workloads. Currently under the testing phase, Rebellions is working with Samsung to produce it on a large scale and launch it later this year. Designed to power large-scale AI systems, REBEL-Quad achieves the highest level of energy efficiency while delivering top performance. Equipped with the latest HBM3E, REBEL-Quad demonstrates outstanding computing speed and throughput per Watt, making state-of-the-art AI services more accessible. REBEL-Quad gives companies the extra edge to stay ahead of the competition.

Rebellions' upcoming semiconductor is being manufactured by Samsung using its advanced 4-nanometer process, one of the most cutting-edge chipmaking technologies available. For comparison, Nvidia’s Blackwell chips also use a 4-nanometer process, though produced by TSMC. The chip will integrate high bandwidth memory (HBM), a stacked memory type essential for managing large-scale data workloads. An HBM vendor has not yet been selected.