Rebellions
Rebellions is a South Korean fabless AI semiconductor company that designs chiplet-based neural processing units (NPUs) for data center AI inference, serving hyperscalers, telecom operators, and enterprises across Korea, the US, Japan, and the MENA region.
- Company typePrivate
- Founded2020
- HeadquartersSeongnam, South Korea
- Headcount251–500
- GTM typeB2B
- OfferingHardware or Manufacturing
What Rebellions does
Rebellions is a fabless AI semiconductor company founded in 2020 and headquartered in Seongnam, South Korea, that designs inference-optimized neural processing units (NPUs) for data center and enterprise AI workloads. The company's product portfolio spans three generations: the first-generation ATOM family (RBLN-CA02/CA22/CA25) manufactured on Samsung 5nm; the second-generation Rebel100, a four-homogeneous-chiplet SoC using UCIe-Advanced interconnects with 144GB HBM3E delivering 2,048 TFLOPS FP8 at up to 600W; and the third-generation Rebel-Quad, an industry-first quad-chiplet NPU achieving 2 FP8 PFLOPS at 600W using four 320mm² Samsung SF4X dies, presented at ISSCC 2026 and demonstrated live at Hot Chips 2025. Rebellions has vertically integrated its silicon into RebelServer (5U, 8x RebelCards delivering 2 PFLOPS FP8), RebelRack, and RebelPOD rack-scale systems, complemented by a full software stack (RBLN SDK) including the RBLN Compiler, Compute Library, Runtime, Driver, Firmware, a PyTorch-native inference engine, the vllm-rbln plugin (with OpenAI-compatible endpoints), and a 300+ model zoo spanning LLMs, generative AI, vision, speech, and physical AI workloads. The company monetizes through direct hardware sales (Rebel100/ATOM chips and RebelServer/RebelRack/RebelPOD rack-scale platforms), bundled 3-year hardware/software support, and vertical-specific solution deployments via implementation partners, with quote-based custom enterprise pricing and no public SKUs. Customers include hyperscalers, telecom operators, governments, and enterprises across Korea, the US, Japan, and the MENA region, with named deployments at SK Telecom, Saudi Aramco, KT Cloud, KB Financial, Mongolian Customs, ECOPEACE (UAE), IGLOO, KOLON BENIT, and Superb AI.
Rebellions has raised $850 million cumulatively across multiple rounds, with its Series C ($250M, September 2025, $1.4B valuation) led by Arm and Samsung Ventures, and its pre-IPO round ($400M, March 2026, ~$2.34B valuation) co-led by Mirae Asset Financial Group and the Korea National Growth Fund under the "K-Nvidia" initiative. The company merged with SK Telecom-backed Sapeon Korea in December 2024 to form Korea's first AI chip unicorn, established subsidiaries in Japan (February 2025), Saudi Arabia (August 2025), and the US (November 2025), and appointed JPMorgan as global lead underwriter for a planned Seoul IPO targeted for late 2026 or early 2027. Channel and OEM partnerships include Tomen Devices, Giga Computing, Pegatron, Red Hat (OpenShift AI), CoAsia SEMI, Marvell, and Credo Technology, alongside strategic alliances with Arm, Samsung Foundry, and SK Hynix for chiplet architecture, HBM co-design, and 2nm AI CPU development.
Rebellions firmographics
Firmographics- Name
- Rebellions
- Legal name
- Rebellions Inc.
- Website
- https://rebellions.ai
- Company type
- Private
- Founded year
- 2020
- Operating status
- Operating
- Headcount range
- 251–500 employees
- Short description
- Rebellions is a South Korean fabless AI semiconductor company that designs chiplet-based neural processing units (NPUs) for data center AI inference, serving hyperscalers, telecom operators, and enterprises across Korea, the US, Japan, and the MENA region.
- Ownership category
- akta.pro rank
Rebellions industry classification
Industry- Product category
- AI Inference Accelerators (NPU Semiconductors)
- NAICS
- Computer Systems Design and Related Services (54151), Computer Systems Design Services (541512)
- SIC
- Services-Computer Integrated Systems Design (7373)
- akta.pro primary industry
- Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC) (HDAHABAC)
- akta.pro secondary industries
- AI Accelerators (GPUs/TPUs/NPUs/ASICs) (HDAAAAAA), AI Hardware IP & Chiplet/Packaging for Accelerators (EDA/IP, advanced packaging) (HDAAAAAL)
Keywords
Where Rebellions is headquartered
LocationHeadquarters
- HQ city
- Seongnam
- HQ country
- South Korea
- HQ region
- Asia
Offices5 records
Markets served
Rebellions business model
Business model- GTM type
- B2B
- Offering type
- Hardware or Manufacturing
- Cost components
- Technology or R&D, Personnel, Supply Chain, Marketing or Sales, Operations, Infrastructure
Revenue model
- Hardware sales (NPU chips and accelerators): Direct sale of NPU silicon products — ATOM (RBLN-CA02, ATOM-Max), Rebel100, and the Rebel-Quad chiplet platform — to enterprise and government customers, with RebelServer 5U platforms shipping with 8 RebelCards each.
- Integrated rack-scale AI infrastructure (RebelRack, RebelPOD): Vertically integrated, fully deployable AI infrastructure platforms built on Rebel100 NPU sold as production-ready data center solutions, extending the offering beyond silicon into rack-scale systems.
- Software and SDK (RBLN SDK): Sale and licensing of RBLN SDK, RBLN Compiler, runtime modules, and developer tooling to enable customers to deploy AI inference on Rebellions hardware with vLLM, PyTorch, Triton, and Hugging Face integration.
- Hardware support and maintenance services: RebelServer ships with 3-year business standard hardware and software support bundled, providing recurring service revenue.
- Vertical-domain AI solution deployments (LegalTech, customs, healthcare, etc.): Domain-specific full-stack AI solutions combining Rebellions NPU with partner AI models (e.g., PlutoLM with Wert Intelligence for patent analytics; Rebellions ATOM with on-premise appliances for IGLOO SOC).
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Other | Multi-year contract | Custom / quote-based enterprise pricing for all Rebel100, RebelServer, RebelRack, RebelPOD, ATOM-Max and software products |
Go-to-market motion5 records
Distribution channels7 records
Marketing channels7 records
Rebellions product offering
Product offeringCore offering
Rebellions designs and sells fabless AI inference NPU chips (Rebel100, ATOM-Max) and vertically integrated rack-scale AI infrastructure systems (RebelServer, RebelRack, RebelPOD) for data center deployments. Its chips use a chiplet-based architecture with UCIe-Advanced interconnects, HBM3E memory, and mixed-precision FP8/FP16 engines, targeted at large language model inference workloads. The company also provides the RBLN SDK software stack for deploying PyTorch, TensorFlow, Hugging Face, and vLLM models on its hardware.
Product overview
Rebellions offers a vertically integrated AI inference platform spanning from silicon to system, purpose-built for energy-efficient, high-performance inference at scale. The portfolio is anchored by the Rebel100™ NPU — a chiplet-based accelerator using UCIe-Advanced and 144GB HBM3E — which is offered as the RebelCard™ module and integrated into the RebelServer™ (5U, 8x RebelCard), the RebelRack™ rack-scale system, and the RebelPOD™ AI cluster. The previous-generation ATOM™-Max chip (RBLN-CA25) is delivered through the ATOM™-Max Server and ATOM™-Max POD systems. All hardware is supported by the Rebellions SDK (RBLN SDK) software stack — including the RBLN Compiler, Runtime, Compute Library, Driver, and Firmware — and the Rebellions Scalable Design (RSD) framework for distributed multi-NPU inference. The 300+ model RBLN Model Zoo, native vLLM/PyTorch/TensorFlow/Hugging Face support, and the vllm-rbln plugin (with OpenAI-compatible API) complete the developer-facing layer.
Differentiator
Problem solved
Functional benefit
Brands
- Rebel100: Chiplet-based AI inference accelerator featuring 4-homogeneous-chiplet SoC with UCIe-Advanced, HBM3E 144GB, up to 2,048 TFLOPS FP8 performance
- RebelServer
- RebelRack
- RebelPOD
- ATOM
- ATOM-Max Server
- ATOM-Max POD
- Rebellions SDK
- RebelCard
Products and services
- Rebel100 Second-generation AI inference NPU featuring a 4-homogeneous-chiplet SoC based on UCIe-Advanced, with 1,024 TFLOPS FP16 / 2,048 TFLOPS FP8 performance, 144GB HBM3E at 4.8TB/s, 16Gbps UCIe-A chiplet interface, 2x PCIe Gen5 x16 host interface, and up to 600W power. Designed for frontier LLM serving for enterprise data centers and hyperscale AI labs.
- RebelCard Module-type card built on the Rebel 100 NPU, designed to deliver GPU-class performance with superior power efficiency for large-scale AI data centers. Integrated into Arm AGI CPU servers and Giga Computing's server platforms.
- RebelServer 5U server system with 8x RebelCard (each powered by a Rebel100 chip) delivering up to 2 PFLOPS FP8 performance. Built on 2x AMD EPYC 9355 CPUs, 1.5TB DDR5 memory, 4x 400G QSFP112-DD networking, 6x 2700W PSU, with typical power draw 4-6kW. Ships with 3-year hardware and software support. Native integration with vLLM, PyTorch, and Triton.
- RebelRack Rack-scale AI inference system built on the Rebel100 NPU, designed to improve data center efficiency by reducing power use and deployment costs. Production-ready for enterprise data centers with Kubernetes and PyTorch integration.
- RebelPOD Fully deployable, vertically integrated AI cluster platform built on the chiplet-based Rebel100 NPU, extending Rebellions' offerings into production-ready data center solutions. Includes Credo ZeroFlap active electrical cables for stable, high-performance connectivity.
- ATOM-Max (RBLN-CA25) Previous-generation AI accelerator delivering 128 TFLOPS FP16 and up to 512 TOPS INT8 / 1024 TOPS INT4, with 1024GB/s bandwidth via 64GB GDDR6, PCIe Gen5 x16 host interface, air cooling (passive), 350W max power, in FHFL dual-slot form factor.
- ATOM-Max Server
Quantifiable outcome
- 2x performance-per-watt vs GPUs in UAE water purification deployment (ECOPEACE) — TTA certified
- +10 more outcomes
Companies that use Rebellions
Customer profileNamed customers12 records
Ideal customer profiles4 records
Rebellions technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
Integration13 records
AI capability14 records
Feature12 records
Rebellions partnerships and signals
Strategic signalPartnerships
30 partnerships are on record, tiered core, flagship / core, minor and major prospects.
- Giga Computing (GIGABYTE subsidiary)coreMOU signed at Computex 2026 in Taipei to jointly develop next-generation AI server and rack-scale solutions based on RebelCard, with a focus on sovereign AI. Combines Rebellions' chip technology with Giga Computing's server platform and global infrastructure network.
- KB Financial GroupcoreStrategic MOU to develop next-generation AI financial infrastructure in Seoul. Rebellions provides AI chip inference technology; KB Financial supports AI use cases and ecosystem. KB Investment participated in Series A (2022) and KB Securities in Series B.
- Credo Technology GroupcoreCollaboration to integrate Credo's ZeroFlap active electrical cables into RebelPOD high-performance AI cluster for enterprise AI inference workloads, eliminating intermittent link disruptions and improving uptime.
- SK Telecomflagship / coreDeploying Rebellions NPU in "A.", Korea's largest LLM service (first large-scale NPU deployment in a real AI application). Strategic MOU to integrate RebelCard into Arm AGI CPU server and validate in SKT's AI Data Center. SKT is also a strategic investor.
- Red HatcoreIntegration of Rebellions' NPUs into Red Hat OpenShift AI platform combined with vLLM inference engine. Rebellions, Red Hat, and Kolon Benit plan joint seminars and consulting. Enables enterprise distribution via Red Hat's open ecosystem.
- Tomen DevicescoreMOU signed to enter Japan's AI inference market leveraging Tomen Devices' distribution network.
- Saudi Aramco (Wa'ed Ventures)flagship / core$14.4M strategic investment from Aramco's venture arm Wa'ed Ventures; Rebellions established a Riyadh MENA HQ to supply AI chips to Aramco's AI data centers; in discussions with Saudi telecom and IT firms.
- Marvell TechnologycoreCollaboration to offer high-performance, energy-efficient AI systems tailored for sovereign-backed AI initiatives across APAC and the Middle East. Rebellions designs customer-specific AI accelerators using Marvell's custom platforms, advanced packaging, high-speed SerDes, and die-to-die interconnects.
- CoAsia SEMIcoreJoint development agreement to develop next-generation AI chiplet and server software solutions for data centers, building on Rebellions' REBEL architecture with CoAsia SEMI's 2.5D silicon interposer and packaging technologies. Mass production planned for US, Europe, and Japan by end of 2026.
- PegatroncoreStrategic partnership on AI hardware using Rebellions' chiplet-based accelerator, announced November 2024.
- Armflagship / coreLong-running strategic alliance including joint AI chiplet technology development (2024), joining Arm Total Design (October 2025), strategic investment by Arm in Rebellions' Series C, and 3-party collaboration with SK Telecom integrating RebelCard into Arm AGI CPU server (April 2026). Arm AGI CPU is co-developed with Meta.
- Samsung Foundryflagship / coreManufacturing and co-design partnership across multiple process nodes. ATOM manufactured on Samsung 5nm; Rebel-Quad on Samsung SF4X; co-design for 2nm AI CPU chiplet with ADTechnology and Arm. Samsung Ventures is also a Series C investor.
- SK Hynixflagship / coreCo-design partnership for custom HBM memory including HBM4e for second-generation Rebel (144GB). Also a Series C investor and key supply chain partner for high-bandwidth memory.
- Sapeon Korea (SK Telecom subsidiary)flagship / coreMerger completed in December 2024 forming Korea's first AI chip unicorn, combining Rebellions' NPU technology with SK Telecom-backed Sapeon. Unified Korean AI semiconductor capabilities and deepened HBM supply chain.
- KT / KT CloudcoreFirst customer shipment of ATOM in May 2023; strategic investor; KT Cloud powers 3.45 PFLOPS computing infrastructure using Rebellions ATOM chip.
- SynopsyscoreEDA toolchain partner; Rebellions used Synopsys VCS, ZeBu, Virtualizer, and Verification Continuum to achieve full bring-up and live demo of Rebel-Quad in just 5 weeks after first silicon, with 95-98% accuracy in predicting actual silicon performance.
- ADTechnologycoreCo-development partner with Arm, Samsung Foundry, and Rebellions on a 2nm AI CPU chiplet platform. Part of Project LeapFrog multi-vendor initiative.
- VESSL AIminorMOU to build cloud infrastructure using domestically developed NPUs, combining Rebellions' NPU hardware with VESSL AI's orchestration platform to target physical AI workloads in robotics and smart manufacturing, plus Korea's public sector and overseas markets in MENA and Southeast Asia.
- OCP (Open Compute Project)coreMember of the Open Compute Project and participant in OCP Global Summit 2025 and OCP EMEA Summit 2026; contributes to chiplet and rack-scale standards for AI data centers.
- KOLON BENITcoreMOU to build AI Vision Intelligence safety monitoring solution for construction sites using hybrid GPU-NPU infrastructure. Deployed at KOLON GLOBAL construction sites with up to 50% lower power consumption and 45% less heat vs GPU servers.
- Wert IntelligencecoreMOU to build NPU-based full-stack solution for patent analytics combining Rebellions' ATOM NPU with Wert Intelligence's PlutoLM AI model and keywert service. Deployed on-premise at L&F for secondary battery cathode material patent analytics.
- Standard EnergyminorPartnership to integrate Rebellions' NPUs (5x more power-efficient than GPUs) with Standard Energy's Vanadium Ion Battery (VIB) ESS for data center power delivery. Joint blueprint for sustainable AI infrastructure worldwide.
- Superb AIminorDeployed Superb AI's object recognition food detection model on Rebellions' NPU-based cloud service, achieving 2x FPS/W vs GPU-based cloud services.
- ECOPEACE (UAE)minorECOBOT autonomous water-purification robot solution in UAE powered by central Rebellions ATOM NPU server; achieved 2x performance-per-watt vs GPUs (TTA certified).
- IGLOO CorporationminorBuilt AiR (AI Road) NPU-based on-premise security appliance for SOCs using Rebellions NPU, achieving 1.6x better performance-per-watt vs GPUs and 50% lower power.
- SK Telecom (X Caliber veterinary imaging)coreSK Telecom's X Caliber AI veterinary imaging service fully migrated inference infrastructure from GPUs to Rebellions' ATOM NPU — first large-scale subscription-based AI service (~1,000 veterinary clinics) operating entirely on NPU-only infrastructure.
- SKT Consortium (Krafton, 42dot, Selectstar, Liner, Seoul National University, KAIST)coreConsortium selected by Korea's Ministry of Science and ICT for the "Proprietary AI Foundation Model" project; Rebellions contributes NPU hardware to develop a national AI model.
- Penguin SolutionsminorStrategic collaboration initiative to advance global AI data center ecosystem (with SK Telecom).
- DOCOMO InnovationscoreThree-way partnership with SK Telecom and DOCOMO Innovations to accelerate next-gen AI infrastructure for telecom operators (April 2025).
- Meta and xAI (target customers)major prospectsRebellions is targeting major U.S. AI labs (Meta and xAI) as customers for its Rebel100 NPU and RebelServer systems; plans to scale production of Rebel100 to meet this demand.
Scale indicators14 records
Recent moves4 records
Expansion highlights8 records
Rebellions competitors and assessment
Company assessmentDirect peers
- Groq: Groq builds LPU inference accelerators purpose-built for low-latency LLM serving, the same core use case as Rebellions' Rebel/ATOM NPUs. Both target production AI inference with an emphasis on deterministic latency and TCO advantages over Nvidia GPUs.
- FuriosaAI: FuriosaAI is Rebellions' closest Korean peer, developing RNGD AI accelerators for LLM inference and selected alongside Rebellions for Korea's K-Perf benchmark. Both compete for Korean sovereign AI customers and government-backed deployments.
- Cerebras Systems: Cerebras builds wafer-scale AI accelerators (WSE) used for both training and inference at hyperscaler and enterprise customers. Comparable to Rebellions as a fabless AI chip startup offering an alternative to Nvidia for large-model workloads.
- SambaNova Systems: SambaNova develops RDU-based AI accelerators (SN40L) targeting enterprise and government LLM inference and training. Competes head-to-head with Rebellions in the inference-optimized AI accelerator category and on enterprise/government RFPs.
- Tenstorrent: Tenstorrent designs RISC-V-based AI accelerators and chiplet-based systems for training and inference, positioning itself as a fabless Nvidia alternative. Comparable to Rebellions in targeting hyperscalers and sovereign AI customers with chiplet-based NPU architecture.
Emerging players
- Hailo: Hailo builds edge-AI inference accelerators (Hailo-8, Hailo-15) used in cameras, automotive, and industrial applications. Adjacent to Rebellions' physical AI / on-prem appliance use cases (e.g., IGLOO, KOLON BENIT, ECOPEACE) but focused on the edge rather than data center.
Broad incumbents
- Marvell Technology: Marvell provides custom-silicon platforms, advanced packaging, high-speed SerDes, and die-to-die interconnects used in hyperscale AI accelerators. Marvell is also a direct collaborator with Rebellions for sovereign AI customers in APAC and MENA, making it both an enabler and a competing custom-silicon option.
- NVIDIA: Nvidia is the dominant GPU provider (H100, H200, B100/B200, GB200) for AI training and inference with the entrenched CUDA software stack. Rebellions explicitly positions its NPUs as a power-efficient Nvidia alternative, and Nvidia's roadmap and pricing directly set the competitive bar.
- AMD: AMD's Instinct MI300/MI325 GPUs and ROCm software target the same AI inference workloads as Rebellions' NPUs, and AMD EPYC CPUs are used inside RebelServer. AMD is a more direct competing accelerator vendor but with a much broader portfolio.
Others
- Sapeon Korea (now part of Rebellions): Sapeon Korea was the SK Telecom-backed AI chip subsidiary that merged into Rebellions in December 2024 to form Korea's first AI chip unicorn. Pre-merger it was a direct Korean NPU peer; the combined entity now consolidates domestic inference-silicon IP and SK Hynix HBM access.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat7 records
Key risks6 records
Key highlights7 records
Customer concentration
Rebellions social profiles
Digital presenceRebellions financial estimates
Financial estimateRevenue estimate
Valuation estimate
Rebellions leadership team
Management profileNumber of profiles
Profiles6 records
Rebellions subsidiaries and ownership
Company hierarchySubsidiaries3 records
Rebellions funding detail
Funding detailFunding overview
Funding rounds8 records
Investors49 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Rebellions M&A and investment
M&A and investmentM&A1 record
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Rebellions
What does Rebellions do?
Rebellions designs and sells fabless AI inference NPU chips (Rebel100, ATOM-Max) and vertically integrated rack-scale AI infrastructure systems (RebelServer, RebelRack, RebelPOD) for data center deployments. Its chips use a chiplet-based architecture with UCIe-Advanced interconnects, HBM3E memory, and mixed-precision FP8/FP16 engines, targeted at large language model inference workloads. The company also provides the RBLN SDK software stack for deploying PyTorch, TensorFlow, Hugging Face, and vLLM models on its hardware.
Is Rebellions a public or private company?
Rebellions is a private company. It is classified as venture growth investor backed and is currently operating.
When was Rebellions founded?
Rebellions was founded in 2020. It employs 251 to 500 people.
Where is Rebellions based?
Rebellions is headquartered in Seongnam, South Korea, in the Asia region.
How does Rebellions make money?
Five revenue lines are on record. Hardware sales (NPU chips and accelerators) is the primary driver. The others are integrated rack-scale AI infrastructure (RebelRack, RebelPOD), software and SDK (RBLN SDK), hardware support and maintenance services and vertical-domain AI solution deployments (LegalTech, customs, healthcare, etc.).
Who are Rebellions's main competitors?
Direct peers on record are Groq, FuriosaAI, Cerebras Systems, SambaNova Systems and Tenstorrent. Hailo is listed as an emerging player. Broad incumbents are Marvell Technology, NVIDIA and AMD. Sapeon Korea (now part of Rebellions) is listed as an others.
Does Rebellions have an API?
Yes. Rebellions offers a Python SDK (RBLN SDK: rebel-compiler, optimum-rbln, vllm-rbln packages) and a C/C++ runtime for compiling and running AI models on Rebellions NPUs. The vllm-rbln plugin implements paged attention and continuous batching on the NPU and exposes an OpenAI-compatible API endpoint for LLM serving, enabling developers to deploy models trained with PyTorch, TensorFlow, or Hugging Face onto Rebellions hardware. SDK installation is via pip (pip3 install --extra-index-url https://pypi.rbln.ai/simple rebel-compiler optimum-rbln vllm-rbln). Developer documentation is at docs.rbln.ai.
What industry is Rebellions in?
Rebellions's product category is AI Inference Accelerators (NPU Semiconductors). Its primary akta.pro industry code is HDAHABAC, Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC), with a secondary code of HDAAAAAA, AI Accelerators (GPUs/TPUs/NPUs/ASICs). Its NAICS code is 54151 and its SIC code is 7373.