News
IndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value ChainIndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value Chain
EveryChip
ATOM

ATOM

AI inference system on a chip by Rebellions

Rebellions designed the ATOM system on a chip for artificial intelligence inference workloads. It serves as the second generation product from the South Korean firm, moving beyond high frequency trading hardware. The chip powers cards with varying power limits and memory configurations for data centre use.

Share on WhatsApp

  • Four chiplets act as one die for inference
  • Each chip has 16 GB of GDDR6 memory
  • Max card delivers 1024 TOPS of INT4 compute
  • Runs SK Telecom A.X K1 sovereign LLM
  • Preceded by ION chip for trading markets

What it is

The ATOM is a system on a chip built for artificial intelligence inference. Rebellions, a company based in South Korea, created it as their second generation silicon product. This design marks a shift from their earlier focus on high frequency trading hardware. The chip integrates acceleration and memory architecture to support large language models and agentic workflows in production environments.

How it came about

Rebellions started by making chips for high frequency trading markets. Their first system on a chip was called ION. The company then transitioned to the artificial intelligence sector with the ATOM design. This second generation product aims to provide energy efficient infrastructure for AI service providers. It has moved from initial silicon to commercial deployment with paying customers.

Inside the chip

The chip uses four chiplets that act as one die. Each ATOM unit contains 16 gigabytes of GDDR6 memory running at 16 gigabits per second. This setup provides 256 gigabytes per second of bandwidth per chip. Compute capabilities include 32 teraflops of FP16 performance. Integer operations reach 128 TOPS for INT8 and 256 TOPS for INT4. The design prioritises inference tasks from the ground up.

How it was made

The chip is packaged onto cards with different power profiles. The largest card configuration holds four ATOM chips. Memory bandwidth totals 1024 gigabytes per second for the full card. The company claims active deployment in global production systems.

Where it went

The chip powers the RebelServer infrastructure. It supports generative AI and agentic workflows for enterprises. SK Telecom used it to run their sovereign large language model called A.X K1. The hardware integrates into existing data centres without requiring new construction. Partnerships extend to Japan for energy efficient inference infrastructure.

Versions and family

Cards come in three main forms. The RBLN-CA21 has a single chip and stays under 75 watts. The RBLN-CA22 also has one chip but draws up to 90 watts. The ATOM Max card holds four chips and consumes 350 watts. Rebellions plans a third generation chip called REBEL which will use HBM3e memory.

Why it matters

The company states it has achieved multi-region commercial deployment. This distinguishes it from startups that only complete infrastructure pilots. Real AI service operators pay for the hardware. The design aims to make advanced models operate economically while remaining under user control.

Questions people ask about ATOM

What is the ATOM chip used for?

It is designed for artificial intelligence inference workloads. The chip helps enterprises and service providers run large language models and agentic workflows efficiently in data centres.

How much memory does each ATOM chip have?

Each individual ATOM chip includes 16 gigabytes of GDDR6 memory. This memory runs at 16 gigabits per second on a 128-bit bus to provide high bandwidth for data processing.

Who designed the ATOM processor?

Rebellions, a South Korean startup, designed the ATOM system on a chip. The company previously focused on high frequency trading hardware before transitioning to AI acceleration products.

What comes after the ATOM chip?

The third generation system on a chip is called REBEL. It will move to HBM3e memory stacks to increase capacity and bandwidth compared to the GDDR6 used in ATOM.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on ATOM yet, so the facts on this page come from the sources listed above: rebellions.ai/, chipsandcheese.com/p/rebellions-from-high-frequency-trading. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryChip wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    Photo credits on this page

    The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

    • Tensor Processing Unit: Norman P. Jouppi, George Kurian, Sheng Li, Peter Ma, Rahul Nagarajan,, CC BY 4.0, source

    The weekly digest

    Every Monday: the chip of the day, new pages on EveryChip, and the conferences and anniversaries of the week ahead.