Challenging NVIDIA from Korea — The Story of Rebellions

"You Don't Need NVIDIA"

In 2022, at a conference hall in Gangnam, Seoul.

Numbers appeared on the screen. AI inference performance benchmarks. Energy efficiency comparison charts. And on the final slide, a single line:

"ATOM delivers up to 3x better energy efficiency than comparable NVIDIA GPUs."

The presenters were two engineers in their early thirties — semiconductor experts who had worked at Samsung Electronics and Qualcomm. They had arrived with a chip they designed themselves, just two years after founding their company.

The company's name: Rebellions.

The name said it all. A rebellion. A challenge to the established order.

A Dream Born at KAIST

South Korea Korea's new challenge in AI semiconductors

Rebellions was co-founded by two people.

Oh Dong-hoon — CEO. Electrical Engineering graduate of KAIST. Formerly designed semiconductor architecture at Qualcomm Korea.

Park Sung-hyun — CTO. Also a KAIST graduate. Previously worked on AI chip development at Samsung Electronics' System LSI division.

In 2020, both were wrestling with the same question.

"The AI market is exploding, but every chip is NVIDIA. Could we build our own?"

At the time, NVIDIA dominated the AI semiconductor market entirely. From data center training to inference, NVIDIA GPUs were everywhere. Google (TPU) and Amazon (Trainium) were building their own chips — but only for internal use.

The two founders saw an opportunity.

"If we build a chip specialized for AI inference, we can beat NVIDIA."

Founding and the First Hurdle

Rebellions was founded in 2020.

The early days were quiet. The concept of a domestic Korean AI semiconductor startup was unfamiliar. Reactions ranged from skeptical to dismissive — "A fabless AI chip startup in Korea?"

Fabless — designing chips without manufacturing them, the same model NVIDIA uses, outsourcing production to foundries like TSMC. But chip design demands enormous capital and expertise. Without proven credentials, raising investment is nearly impossible.

In those early days, CEO Oh Dong-hoon made the rounds pitching investors.

The responses were cold. "Semiconductors require too much capital." "You're not a conglomerate — you can't build an AI chip." "How do you beat NVIDIA?"

But their backgrounds spoke for themselves. Qualcomm and Samsung Electronics — two of the world's elite semiconductor companies. The technical credibility was there.

In 2021, they closed their seed round.

ATOM — The Embodiment of the Challenge

In the second half of 2022, Rebellions unveiled its first AI chip.

Name: ATOM.

The name carries a philosophy. The atom — the fundamental unit of everything. A commitment to perfectly optimizing the most basic building blocks of AI computation.

ATOM's core focus was AI inference optimization.

AI workloads break down into two phases. Training: where the AI model learns from data. Inference: where the trained model delivers answers in live service.

NVIDIA GPUs are optimized for training. But inference is a different game. Inference runs 24/7/365, and energy efficiency matters enormously. Rebellions targeted this gap.

In public benchmarks, ATOM delivered impressive results. Equivalent inference performance to NVIDIA A100 at roughly half the energy consumption. For cloud operators, the cost savings potential was substantial.

Investors Came Running

After ATOM's debut, the atmosphere changed.

2022 Series A — KT (Korea Telecom), KB Investment, Mirae Asset Venture Investment, and others participated. Telecoms and financial investors interested in AI infrastructure came to the table.

2023 Series B — the stakes grew. Samsung Catalyst Fund joined the round. Samsung investing directly was a signal that technical validation was complete. Cumulative investment reached hundreds of billions of Korean won.

As the global AI semiconductor market heated up following ChatGPT, demand for AI inference exploded — and data center operators wanted to reduce their dependence on NVIDIA.

"We want to break free from NVIDIA's monopoly. If there's an alternative that works, we'll use it."

That was why investors were paying attention to Rebellions.

The Merger with SAPEON — The Game Changes

Semiconductor chip The age of AI semiconductor competition

In 2024, a major announcement shook Korea's AI semiconductor world.

Rebellions and SAPEON were merging.

SAPEON was an AI semiconductor subsidiary created by SK Telecom — established to develop AI chips for SK Telecom's own data centers, with its own AI inference chip technology.

The merger was more than a standard M&A deal.

Rebellions' strengths: The agility of a pure-play fabless startup, the technology behind ATOM, the ability to develop external markets.
SAPEON's strengths: The backing of SK Telecom, a large conglomerate, a stable internal customer base (SK Telecom's data centers), and deep capital reserves.

Post-merger, the combined company kept the name Rebellions. The Rebellions brand had stronger recognition.

The goal of the merger was clear: "Build the scale and technology needed to survive in the global AI semiconductor market."

Why This Fight Matters

The AI semiconductor market is projected to grow to tens of trillions of won by 2030.

NVIDIA dominates it. H100, H200, B100 — NVIDIA keeps releasing new chips and widening its lead.

But not every country or company is comfortable with that level of dependence.

The US has begun restricting AI chip exports. China is developing its own AI chips to operate without NVIDIA. Europe is calling for AI semiconductor self-sufficiency.

Korea is no different. Being entirely reliant on a US company for AI infrastructure is strategically vulnerable. Samsung Electronics and SK Hynix lead the world in memory, but in AI chip design, Korea's presence has been limited.

Rebellions is an attempt to fill that gap.

The Philosophy Behind Rebellions

In an interview, CEO Oh Dong-hoon put it plainly:

"We're not trying to follow NVIDIA. We're trying to be the best in the areas where NVIDIA isn't. AI inference, energy efficiency, specialized workloads — we can win there."

"The goal is to have a Korean AI chip running in data centers around the world."

Rebellions' strategy isn't frontal assault. It's finding the gaps — carving out territory in markets NVIDIA can't serve equally well: inference-specialized workloads, edge AI, energy efficiency. Building from there.

Rebellions Today

After the merger with SAPEON, Rebellions became a larger company.

Next-generation chip development is underway. The successor to ATOM targets higher performance and greater efficiency — with a focus on efficient inference for large language models (LLMs) at the scale of GPT-4 and beyond.

Customer acquisition is ongoing. Domestic telecoms, financial institutions, and government agencies are increasingly adopting ATOM-based systems.

The challenges remain significant. NVIDIA releases ever more powerful chips every year. In terms of capital, ecosystem, and brand, the gap is immense. A meaningful global market share for Rebellions is still a long road ahead.

But two things are certain.

First, the technology is real. Samsung Catalyst Fund investing and SAPEON choosing merger are both proof of that.

Second, the market wants an alternative. If NVIDIA's monopoly cannot last forever, the company that fills that space is being built right now.

The Rebellion Continues

Rebellions. The rebels.

When two engineers — alumni of Qualcomm and Samsung — declared "we will build AI chips," many called it reckless.

But they built ATOM. Investors followed. SAPEON chose to merge. And today, Rebellions is the name that represents Korean AI semiconductors.

The infrastructure war of the AI era has already begun. On that stage, dominated by NVIDIA, a band of rebels from Seoul has taken their place.

How far they can go is still being written.