Rebellions

Founded in September 2020 by ex-Morgan Stanley/SpaceX quant Park Sung-hyun and ex-IBM chip designer Oh Jin-wook, Rebellions designs fabless AI accelerators targeting data-center LLM inference. Its ATOM/ATOM Max NPUs entered mass production in 2023 via Samsung Foundry; the next-generation REBEL NPU (REBEL-Quad chiplet, 144 GB HBM3E) targets GPU-class compute at NPU-class efficiency. In June 2024, Rebellions and Sapeon Korea — SK Telecom's internal AI-chip unit — announced a definitive merger agreement on terms valuing the combined entity at roughly ₩1.3 trillion (~$929M). Sapeon Korea was the surviving legal entity, but Rebellions' leadership and branding carried through; the deal closed December 2, 2024, and the merged company trades as Rebellions. SK Telecom and SK Hynix became strategic investors alongside Samsung. A $250M Series C in September 2025 (Arm and Samsung anchored) valued the company at $1.4B. In March 2026, Rebellions closed a $400M pre-IPO round led by Mirae Asset Financial Group and the Korea National Growth Fund, pushing valuation to $2.34B and total capital raised to $850M. Alongside the round, the company launched RebelRack™ and RebelPOD™ — rack- and pod-scale AI infrastructure products built on the Rebel100™ accelerator. An IPO on the Korean exchange is targeted for late 2026 or early 2027.

Rebellions is the leading independent AI chip challenger out of South Korea — a country that supplies the memory (SK Hynix HBM) and packaging underpinning global AI compute but has lacked a domestic inference-chip champion. The Sapeon merger created the first Korean AI chip unicorn and locked in Samsung, SK Hynix, and Arm as investors; the pre-IPO round at $2.34B signals serious international investor conviction. If its REBEL-Quad NPU hits performance targets in mass production, Rebellions is positioned as a credible regional alternative to Nvidia for inference workloads across Asia.

rebellions.ai AI Accelerators

South Koreaprivate

At a glance

Total raised

$850M

last round pre-ipo

Headcount

—

no public estimate on record

Headquarters

Seoul

South Korea

Founded

2020

6 years active

Status

Private

AI Accelerators

Latest valuation

$2.3B

Pre-IPO · 2026

For subscribers

Subscribe to see the full profile

Including Funding rounds, Bull / Bear thesis, Stock + earnings, Roster changes, Patents, News, and Open roles.

Founding

50% off

/month

91 of 100 spots left

Full

$10

/month

Available once all 100 Founding Member spots are claimed.

Get full access

Already subscribed? Sign in →

Funding history

Total raised

$850M

Rounds tracked

Latest stage

Pre-IPO

Latest valuation

$2.3B

Date	Stage	Amount	Valuation	Lead investors
Mar 30, 2026	Pre-IPO	$400M	$2.3B	Mirae Asset Financial Group, Korea National Growth Fund
Sep 30, 2025	Series C	$250M	$1.4B	Arm, Samsung Ventures, Pegatron VC, Korea Development Bank, Korelya Capital, Lion X Ventures
Jul 1, 2024	Series B Extension	$15M	—	Wa'ed Ventures (Saudi Aramco)
Jan 30, 2024	Series B	$124M	$650M	KT, Pavilion Capital (Temasek), Korea Development Bank, Korelya Capital, DG Daiwa Ventures
Jul 5, 2022	Series A Extension	$22.8M	—	KT
Jun 2, 2022	Series A	$50M	—	Pavilion Capital (Temasek), Korea Development Bank, Mirae Asset Venture Investment, SV Investment, IMM Investment, KB Investment, KT Investment, Mirae Asset Capital
Jul 1, 2021	Pre-Series A	$12.4M	—	Kakao Ventures, Shinhan Capital, GU Equity Partners, Seoul Techno Holdings
Aug 1, 2020	Seed	$4.7M	—	GU Investment, Seoul National University Technology Holdings

How they make money

Hardware + recurring servicesPattern

Rebellions is a fabless semiconductor company that derives revenue from selling AI accelerator hardware — chips, server cards, and increasingly full rack- and pod-scale systems. Its ATOM and ATOM Max NPUs are sold as discrete accelerator cards manufactured by Samsung Foundry (4 nm / 5 nm), and its RebelRack and RebelPOD platforms package those chips into complete inference infrastructure units that hyperscalers and enterprises deploy as turnkey data-center nodes. The hardware ASP rises substantially as the company moves up the stack: a single RebelRack ships with 32 REBEL-Quad accelerators and 64 petaFLOPS of FP8 compute, commanding a system-level price point. The recurring layer sits in software and services: Rebellions offers an NPU-as-a-Service (NPUaaS) model through its KT Cloud partnership, where customers consume inference compute via a cloud API rather than buying hardware outright. The software stack — validated on Red Hat OpenShift and integrated with vLLM, PyTorch, Triton, and Hugging Face — generates ongoing support, optimization, and platform contracts. Commercial deployments at SK Telecom (A-dot call recording) and KT Cloud (3.45-petaflop shared inference cluster) are the clearest examples of this hybrid model in practice. Revenue is not publicly disclosed, but the company has confirmed a growing commercial pipeline beyond Korea, with customers in Japan, Saudi Arabia, and the US.

Generated by editorial-sourced · June 4, 2026Analytical summary · not investment advice

Founder backgrounds

Park Sung-hyun (Sunghyun Park)

Co-founder & CEO

Park Sung-hyun earned a bachelor's degree in electronic engineering from KAIST, then a master's and Ph.D. from MIT. He began his career as a Senior Research Scientist at Intel Labs in 2014, working on IoT x86 processor design, before moving to Samsung Research America as a Staff Engineer in 2016. He joined SpaceX in 2017 as a Starlink ASIC Design Engineer, then worked as a quantitative developer at Morgan Stanley before co-founding Rebellions in September 2020. Under his leadership, Rebellions merged with SK Telecom's Sapeon Korea in December 2024 to become Korea's first AI chip unicorn.

Oh Jin-wook

Co-founder & CTO

Oh Jin-wook earned his bachelor's degree in electrical engineering from Seoul National University, then pursued advanced degrees at KAIST, where he completed a Ph.D. in electrical engineering. During his doctoral work he collaborated with Microsoft and Texas Instruments on approximate computing and coarse-grained reconfigurable arrays. He subsequently joined the IBM TJ Watson Research Center as a chip architect, logic designer, and logic power lead, contributing to multiple AI chip R&D projects focused on neural network accelerators. He co-founded Rebellions in 2020 and has overseen the architecture and development of the ATOM, ATOM Max, and REBEL NPU generations.

Kim Hyo-eun

Co-founder & CPO

Kim Hyo-eun holds a B.S. and Ph.D. in electrical engineering from KAIST. She began her career as an engineer at MAXWAVE in 2004, then joined Samsung Electronics as a Senior Engineer in 2013. In 2014 she moved to the AI medical imaging startup Lunit, where she rose to CPO by 2016, gaining experience translating deep learning research into commercial products. She co-founded Rebellions in 2020, where she leads product strategy and roadmap across the ATOM and REBEL NPU families and the RebelRack and RebelPOD system platforms.

Shin Sung-ho

Co-founder

Shin Sung-ho is an AI and algorithms researcher who studied at Seoul National University. He co-founded Rebellions in September 2020 alongside Park Sung-hyun, Oh Jin-wook, and Kim Hyo-eun. His research background in AI algorithms contributed to the theoretical foundations underpinning Rebellions' NPU architecture. Detailed public information about his specific role and prior career history is limited.

AI bull / bear

Bull case

Rebellions enters its IPO window as the only AI chip company in the world with Samsung Foundry, SK Hynix (HBM), and Arm simultaneously invested and supplying it — a supply-chain moat that is structurally difficult to replicate. Its ATOM NPU has three-plus years of live inference deployment behind it at SK Telecom and KT Cloud, giving it a legitimacy advantage that most GPU-challenger paper-chips lack. The pivot from chip-only sales to rack- and pod-scale systems (RebelRack, RebelPOD) follows Nvidia's own playbook and expands revenue per customer by an order of magnitude; early deployments in Japan, Saudi Arabia, and the US show the TAM is not merely domestic. At $2.34B pre-IPO valuation, the company is priced at a fraction of Nvidia's revenue multiples, and any REBEL-Quad performance validation against H100/H200 in MLPerf would be a material re-rating catalyst.

Bear case

Rebellions has never disclosed revenue figures, and its commercial traction to date is concentrated in two Korean telcos (SK Telecom and KT Cloud) — raising questions about whether it can win workloads against Nvidia in internationally competitive procurement. The REBEL-Quad chiplet architecture is ambitious but unproven in mass production; chiplet yield challenges have derailed larger and better-funded competitors. The Sapeon merger, while giving strategic backing from SK Telecom, also introduced integration complexity: a new CEO-survivor company, two engineering cultures, and overlapping chip roadmaps had to be rationalized in under 18 months. Finally, the planned Korean IPO is a liquidity event in a relatively illiquid domestic market; Korean retail investor sentiment could pressure the valuation downward, and the post-IPO trading multiple will be set by a market where infrastructure semiconductors are under-followed.

Generated by editorial-sourced · June 4, 2026Refreshes quarterly · see disclosures

In the news

Rebellions · last 30 days

No articles ingested yet for Rebellions. Once the hourly news pipeline is live, every article the classifier tags as mentioning this company appears here with its one-line AI summary and sentiment.

Watch

Leadership

Park Sung-hyun (Sunghyun Park)

Co-founder & CEO & CEO & Co-founder

—

Oh Jin-wook

Co-founder & CTO & CTO & Co-founder

—

Kim Hyo-eun

Co-founder & CPO & CPO & Co-founder

—

Shin Sung-ho

Co-founder

—

Marshall Choy

Chief Business Officer (appointed Nov 2025)

—

Products

ATOM / ATOM Max NPU
Rebellions' first-generation neural processing unit, manufactured by Samsung Foundry on a 4 nm process and entered mass production in 2023. Targets data-center LLM inference and reached competitive throughput-per-watt scores in MLPerf benchmarks. ATOM Max is the higher-capacity variant. The chip is deployed commercially and was validated on Red Hat AI in 2026, making it the first Korean NPU certified on a major enterprise AI platform.
REBEL NPU / REBEL-Quad
Next-generation chiplet architecture NPU co-developed with Samsung, featuring 144 GB HBM3E memory and targeting GPU-class FP8 throughput at NPU-class power efficiency. The REBEL-Quad bundles four REBEL chiplets into a single module. Mass production was targeted for 2025. It is the compute engine inside the RebelRack and RebelPOD infrastructure platforms.
RebelRack
A rack-scale AI inference system launched in March 2026 alongside the pre-IPO funding close. Features four nodes, 32 REBEL-Quad accelerators, and 64 petaFLOPS of FP8 compute in a single rack, with 800 Gbps Ethernet interconnect. Designed to let hyperscalers and enterprises deploy Rebellions silicon as a drop-in data-center unit without custom integration work.
RebelPOD
A pod-scale AI infrastructure cluster announced in March 2026 that aggregates 8 to 128 RebelRack nodes (each with eight Rebel100 accelerators) into a unified inference fabric. Targets large-model serving at the cluster level and represents Rebellions' move up the stack from chip to full-system AI infrastructure, competing with Nvidia DGX SuperPOD-style offerings.