Rebellions

Rebellions is a South Korean fabless semiconductor company focused on artificial-intelligence inference. Founded in September 2020 and headquartered in Seongnam, the company develops neural processing units (NPUs), compilers, runtime software, accelerator cards, servers, and rack-scale systems for data centers and cloud providers. Its central proposition is to make the deployment of large language models and other AI applications more power-efficient and economically practical than using conventional graphics processing units designed primarily for general-purpose computation or model training. The company was established by a team that included chief executive Sunghyun Park, chief technology officer Jinwook Oh, and chief product officer Hyo-Eun Kim. Park’s experience in electronic engineering, semiconductor design, financial technology, and AI systems influenced Rebellions’ early interest in specialized processors for high-frequency trading and other demanding workloads. As generative AI and large language models became more prominent, the company broadened its focus toward general-purpose inference acceleration. Its decision to build the business in South Korea reflected the country’s semiconductor manufacturing base, engineering talent, and relationships with major technology and telecommunications companies. Rebellions’ first major processor family is ATOM, an NPU designed for inference with an emphasis on performance per watt. The company has also described ATOM-Max, a multi-chip accelerator card, as a way to scale the first-generation architecture for data-center deployments. Its subsequent Rebel100 platform targets larger language models and multimodal workloads. The platform uses a chiplet-based architecture, high-bandwidth memory, and high-speed interconnects, and is intended to be delivered through accelerator cards, servers, and rack-level systems rather than as a bare semiconductor alone. Rebellions’ software strategy includes support for widely used inference technologies such as vLLM and Triton, helping customers migrate workloads from GPU-based environments. The company has pursued a full-stack model combining silicon, compilers, system software, and deployment infrastructure. This approach is intended to reduce the integration burden for cloud operators and enterprise customers while improving utilization and lowering electricity and operating costs. KT Cloud became the first South Korean cloud provider to offer public NPU-based services using Rebellions’ ATOM processors. SK Telecom has used Rebellions hardware in AI services, including call summarization on its A. platform, and has evaluated the company’s products for its NPU farm. LG Electronics is also identified as a customer or user of Rebellions technology. Rebellions expanded its strategic position through investment from South Korean telecommunications and semiconductor companies, including KT, SK Telecom, SK Hynix, and Samsung Electronics. International investors have included Arm, Saudi Aramco’s Wa’ed Ventures, Japan’s DG Daiwa Ventures, and France’s Korelya Capital. In December 2024, Rebellions merged with Sapeon Korea, the AI-chip subsidiary associated with SK Telecom. The combination strengthened the company’s position in the South Korean AI-semiconductor sector and was described as creating the country’s first AI-chip unicorn. Rebellions remains a privately held company and is positioning itself as a domestic and international alternative for inference infrastructure, particularly as data-center operators seek lower power consumption and more specialized hardware.

Rebellions was founded in September 2020 in South Korea by a group of semiconductor and AI engineers, including Sunghyun Park, Jinwook Oh, and Hyo-Eun Kim. Park studied electronic engineering at KAIST and later completed doctoral work at MIT. His professional background included Intel Labs, Samsung in the United States, SpaceX, and Morgan Stanley, where he worked on AI-based systems for high-frequency trading. The original business concept centered on specialized processors for financial applications, but the company later expanded toward broader AI inference as large language models and generative AI accelerated demand for dedicated computing infrastructure. Chief technology officer Jinwook Oh brought semiconductor architecture experience from KAIST and IBM’s T.J. Watson Research Center. The founders chose South Korea rather than the United States in order to work within the country’s semiconductor ecosystem and access its engineering workforce. Rebellions adopted a fabless model, designing chips while relying on external manufacturing and packaging partners. Its technology strategy combined hardware with compilers and system software, reflecting the difficulty of deploying specialized AI accelerators without compatible programming and inference tools. ATOM became the company’s first-generation NPU. Built for inference rather than general-purpose graphics processing, it emphasized lower power consumption and performance efficiency. The chip was manufactured using Samsung’s 5-nanometer process and was offered as part of broader infrastructure deployments. ATOM-Max extended the concept through a multi-chip PCIe card intended for data-center use. Rebellions later developed Rebel100, a higher-performance, chiplet-based platform aimed at large language model and multimodal inference. The platform incorporates high-bandwidth memory, on-chip SRAM, and high-speed die-to-die connectivity, and is planned for deployment through accelerator cards, servers, and rack-scale systems. Commercial adoption began to develop in South Korea’s cloud and telecommunications sectors. KT Cloud became the first South Korean cloud provider to offer public NPU-based services using Rebellions’ ATOM processors. SK Telecom used Rebellions hardware for AI features including call summarization on its A. service and evaluated the processors for a larger NPU farm. LG Electronics was also identified among the company’s customers. These deployments gave Rebellions reference use cases beyond laboratory benchmarks and supported its argument that inference-specific silicon can reduce the power and cost of serving AI models. Funding and strategic partnerships became central to the company’s expansion. KT led a reported Series B round in January 2024. Rebellions also attracted capital from SK Telecom, SK Hynix, Samsung Electronics, Arm, Wa’ed Ventures, DG Daiwa Ventures, and Korelya Capital. In December 2024, it merged with Sapeon Korea, an AI-chip company associated with SK Telecom. The merger brought together two South Korean competitors and strengthened Rebellions’ scale, technology base, and strategic relationships. It also contributed to the company being described as South Korea’s first AI-chip unicorn. By 2025 and 2026, Rebellions was pursuing larger financing rounds and international growth while preparing for a possible public offering. Reported financing included a Series C round in 2025 and a pre-IPO round in 2026. These figures and valuation reports concern private-company financing and should be understood in their stated transaction context rather than as public-market capitalization. Rebellions’ continuing strategy is to provide complete inference infrastructure—chips, software, cards, servers, and rack-scale systems—to cloud operators, data centers, and enterprises seeking alternatives to conventional GPU platforms.

• 2026

  Pre-IPO financing and National Growth Fund investment

  [1]Rebellions (company)en.wikipedia.org

  Rebellions is reported to complete a 640 billion won pre-IPO round and become the first direct-investment recipient of South Korea’s National Growth Fund under the K-NVIDIA initiative.

• 2025

  Series C financing

  [1]Rebellions (company)en.wikipedia.org

  The company raises a reported US$250 million Series C round at a reported valuation of approximately US$1.4 billion.

• 2024

  Series B financing led by KT

  [1]Rebellions (company)en.wikipedia.org

  Rebellions raises a reported US$124 million Series B round led by KT.

• 2024

  Merger with Sapeon Korea

  [1]Rebellions (company)en.wikipedia.org

  Rebellions combines with Sapeon Korea, the AI-chip subsidiary associated with SK Telecom.

• 2020

  Rebellions is founded

  [1]Rebellions (company)en.wikipedia.org

  Rebellions is established in South Korea as a fabless semiconductor company focused initially on specialized AI processing.

A full-stack AI inference infrastructure company emphasizing energy efficiency, performance per watt, and lower operating costs for cloud and data-center customers.

• AI inference NPU

  ATOM

  ATOM is Rebellions’ first-generation neural processing unit for AI inference. It is designed to prioritize energy efficiency and data-center serving workloads rather than the broad graphics and training workloads associated with conventional GPUs. The chip is manufactured on Samsung’s 5-nanometer process and provides dedicated compute and memory resources for deploying language models and other AI applications. Rebellions delivers ATOM as part of a wider hardware and software platform.

• Multi-chip PCIe accelerator

  ATOM-Max

  ATOM-Max is a multi-chip PCIe accelerator card based on the ATOM architecture. By combining multiple ATOM dies with a controller, it is intended to provide higher aggregate inference throughput, memory capacity, and bandwidth than an individual chip while retaining the platform’s emphasis on power efficiency. The product is aimed at data-center and cloud deployments where accelerator cards can be installed in existing server infrastructure.

• High-performance AI inference NPU

  Rebel100

  Rebel100 is a second-generation inference accelerator designed for large language models, multimodal systems, and other large-scale AI workloads. Its architecture uses multiple chiplets, high-bandwidth HBM3e memory, on-chip SRAM, and UCIe-Advanced connectivity. Rebellions positions it as an inference-first alternative to training-oriented GPU platforms, with support for demanding model structures including mixture-of-experts systems. The platform is intended to support cards, servers, and rack-scale deployments.

• AI accelerator card

  RebelCard

  RebelCard is the accelerator-card form factor associated with the Rebel100 platform. It packages Rebellions’ inference silicon for installation in data-center servers and is intended to provide a practical deployment path for customers adopting dedicated AI inference hardware.

• AI inference server

  RebelServer

  RebelServer is a server-level implementation of Rebellions’ inference platform. It combines accelerator hardware and supporting system components so that cloud and enterprise users can deploy AI-serving capacity without assembling an entire stack from individual components.

• Rack-scale AI infrastructure

  RebelRack and RebelPOD

  RebelRack and RebelPOD are planned rack-scale and pod-level extensions of the Rebel100 platform. They are intended to connect multiple inference accelerators into larger systems for data-center operators serving substantial language-model and multimodal workloads. The products reflect Rebellions’ strategy of selling complete infrastructure rather than isolated chips.

• 2024

  Merger with Sapeon KoreaM&A

  Background

  Both Rebellions and Sapeon Korea were developing AI accelerators in South Korea, while the market required greater scale and resources to compete internationally.

  Action taken

  Rebellions merged with Sapeon Korea in December 2024.

  Aftermath

  The transaction consolidated South Korean AI-chip capabilities and strengthened Rebellions’ position with telecommunications and semiconductor-sector partners.

  Source

• 2024

  Expansion from chips to full-stack inference infrastructureStrategy

  Background

  Specialized accelerators require compatible compilers, runtimes, servers, and deployment systems to compete with established GPU ecosystems.

  Action taken

  Rebellions developed a broader offering spanning NPUs, software, PCIe cards, servers, and planned rack-scale systems.

  Aftermath

  The company positioned itself as a complete inference-platform provider for cloud and data-center customers.

  Source

• 2020

  Founding in South KoreaStrategy

  Background

  The founding team considered establishing the company in the United States but identified South Korea’s semiconductor ecosystem and engineering talent as strategic advantages.

  Action taken

  Rebellions was established in South Korea as a fabless AI-semiconductor company.

  Aftermath

  The decision enabled relationships with South Korean chipmakers, telecommunications companies, cloud providers, and strategic investors.

  Source

1. 1Rebellions (company)en.wikipedia.org
2. 2Rebellions official websiterebellions.ai