Rebellions Debuts REBEL-Quad at Hot Chips 2025, breaking AI’s Energy Tax with High-Performance Chiplet Innovation

Palo Alto, California, August 27th, 2025 – Rebellions Inc., an APAC front-runner in AI inference solutions, presents the first look of its next-generation AI accelerator, REBEL-Quad, at Hot Chips Symposium 2025. This marks the company’s biggest step yet in powering frontier AI models at hyperscale with Blackwell-grade performance and superior energy efficiency.

Following the first commercial success of its first-generation product ATOM™ across Saudi Arabia, Japan, and the U.S., Rebellions — the sole AI chip unicorn in APAC, backed by Korea’s largest industry leaders including SK hynix, SK Telecom, KT(Korea Telecom) and Samsung Ventures & Securities — is expanding its product portfolio with REBEL-Quad to meet rising demand for massive-scale AI workloads.

Delivering Blackwell-grade performance with significantly higher energy efficiency, REBEL-Quad supports today’s largest and most dynamic AI models while keeping deployment simple for developers. Built on Samsung’s 4nm process, it is optimized for frontier-scale AI models, including peta-scale Mixture of Experts (MoE) architectures, a cutting-edge technique that routes different parts of a problem to specialized “experts”. Rebellions highlighted this capability with a Qwen3 235B MoE model demonstration at the event.

REBEL-Quad introduces several breakthroughs:

• Energy efficiency & performance: Higher compute density and utilization deliver significantly better performance per watt
• Chiplet interconnect: World’s first adoption of UCIe-Advanced, enabling power-efficient data transfer across chiplets
• HBM3E integration: 144GB HBM3E with 4.8TB/s bandwidth, enough for tens of billions of parameters on a single chip

REBEL-Quad is built on a chiplet-based architecture that scales seamlessly to hyperscale AI workloads. It is the world’s first AI accelerator to adopt UCIe-Advanced, enabling fast and efficient data transfer across chiplets. This design delivers high performance and energy efficiency without compromising latency or coherence, while supporting modular upgrades and long-term adaptability. Rebellions is also extending this chiplet roadmap with upcoming products such as REBEL-IO and REBEL-CPU, aimed at trillion-parameter models and multi-node deployments, with proof-of-concepts with global customers expected soon.

“The AI industry can’t afford to scale on GPUs alone. With REBEL-Quad, we’re showing there is another path — one that delivers GPU-class performance without the energy tax” said Sunghyun Park, CEO of Rebellions “This is our answer to the challenges of serving large AI models: a chip that gives developers the same experience they expect, but at a fraction of the energy.”

The launch is supported by key partners, including Samsung Electronics and Alphawave Semi, who emphasized the significance of this achievement:

“We are proud to support Rebellions with our 4nm process and advanced packaging technologies in bringing REBEL-Quad to life. This collaboration highlights how Samsung Foundry’s advanced manufacturing enables next-generation AI accelerators to achieve exceptional power efficiency at hyperscale,” said Mijung Noh, Vice President of Foundry Technology Planning Team at Samsung Electronics.

“REBEL-Quad marks a significant milestone in the evolution of AI silicon, and Alphawave Semi is honored that our complete UCIe IP solutions are at the core of this industry-first production deployment. Our silicon-proven technology delivers exceptional bandwidth density and ultra-low latency, enabling seamless chiplet integration with the highest levels of reliability. This achievement goes beyond performance—it establishes a robust foundation for scalable and open innovation in AI acceleration,” said Letizia Giuliano, Vice President of Product Marketing at Alphawave Semi. “Rebellions is not only advancing the state of the art but also setting a new standard for the future of high-performance computing.”

About Rebellions

Rebellions develops AI accelerators optimized for Large Language Models and large-scale inference, delivering industry-leading energy efficiency. Its flagship REBEL chip uses chiplet architecture and 144GB HBM3E memory for massive-scale AI inference. The platform includes software for seamless datacenter integration. Building on ATOM’s proven mass-production experience since 2024, Rebellions is backed by SK Telecom, SK hynix, Aramco’s Wa’ed Ventures, and KT. Strengthened by its merger with SK SAPEON, Rebellions is now taking the lead in APAC and setting the stage for global AI semiconductor leadership.

To learn more about Rebellions: https://rebellions.ai/