Top 10 AI Companies in South Korea (2026)

South Korea doesn’t want to rent its intelligence. It wants to own it.

For years, the country was known for building the hardware the world’s AI runs on: memory chips, phones, displays. The models came mostly from California. That’s changing fast. Seoul runs a government contest to pick a home-grown foundation model, chip start-ups raise money at multi-billion-dollar valuations, and a major search portal now belongs to an AI start-up.

On one side sit conglomerates and internet giants with deep pockets and data centres. On the other are specialists that have gone global in one narrow area, whether that’s inference silicon, cancer screening or video understanding. Both are being tested hard. Naver, operator of the country’s top search engine, was dropped from the national model programme in January 2026, while a 2020 start-up made the final three.

That churn is a reason to look closely rather than default to the biggest logo. This guide looks at the AI companies in South Korea that stand out in 2026, what each one sells, and which kind of buyer each suits best.

Why Businesses Are Investing in Artificial Intelligence in South Korea

Korean AI has moved from labs to products companies can buy.

Models built for Korean

Global models handle Korean, but home-grown ones are built around Korean text, law and business context. For banks, public bodies and media groups, that can mean better handling of local documents and a simpler conversation with regulators about where data goes.

A national push behind sovereign AI

The Ministry of Science and ICT runs an independent AI foundation model project that funds and ranks competing teams in stages. Each second-round team was allotted support for 1,000 Nvidia B200 GPUs, with a rental budget of about 40 billion won per team. That backing makes it likelier the leading models will be maintained.

Home-grown AI chips

Korean AI chip companies such as FuriosaAI and Rebellions now ship inference hardware commercially. That gives data-centre operators an alternative to the GPU queue, and several Korean models already run on them.

Proven exports

Some Korean firms earn most of their revenue abroad. Medical AI group Lunit generated about 95% of its first-half 2026 sales outside Korea, proof its tools pass foreign regulators and procurement.

Top 10 AI Companies in South Korea at a Glance

Rank	Company	Primary Strength	Best For	HQ / Founded
1	Naver	HyperCLOVA X models and AI cloud infrastructure	Enterprises wanting Korean-hosted AI	Seongnam / 1999
2	LG AI Research	EXAONE and K-EXAONE foundation models	Industrial and research-heavy organisations	Seoul / 2020
3	SK Telecom	A.X K1, a 519-billion-parameter open model	Telecoms, defence and large deployments	Seoul / 1984
4	Samsung SDS	Enterprise AI platforms and cloud	Large firms and public bodies	Seoul / 1985
5	Upstage	Solar LLMs plus the Daum portal	Finance, legal and document-heavy work	South Korea / 2020
6	FuriosaAI	RNGD inference chips	Data centres cutting inference power costs	Seoul / 2017
7	Rebellions	AI inference servers and racks	Telcos, cloud and neocloud operators	Seongnam / 2020
8	Kakao	Consumer AI inside KakaoTalk	Brands reaching Korean consumers	Jeju / 2006
9	Lunit	AI for cancer screening and oncology	Hospitals, screening programmes, pharma	Seoul / 2013
10	TwelveLabs	Video understanding models	Media, sports and archive owners	San Francisco and Seoul / 2021

Top 10 AI Companies in South Korea (2026)

Ranked on technology, traction and 2025-2026 milestones.

1. Naver

Founded in 1999 and headquartered in Seongnam, Naver runs South Korea’s top search engine and a major cloud business. Its large language model, HyperCLOVA X, launched in August 2023 alongside the CLOVA X chatbot and the Cue: AI search service. In June 2026, Naver and Nvidia announced AI factories on Nvidia’s DSX platform, starting with a 55-megawatt expansion at Naver’s GAK Sejong data centre and planned to scale to gigawatt level.

Naver’s strength is breadth: data, search, cloud and models under one roof. It’s the first Korean member of Nvidia’s Nemotron Coalition and is using that work in its next HyperCLOVA X models. It did take a hit in January 2026, when the science ministry dropped its team from the sovereign model project over the use of an external model’s encoder weights.

Key Solutions:

• ✔ HyperCLOVA X large language models
• ✔ Naver Cloud Platform and AI data centres
• ✔ AI Agent Platform for Korean enterprises

Best for: Enterprises and public bodies that want Korean-built models in Korean data centres.

2. LG AI Research

LG launched its group-wide AI think tank in December 2020. LG AI Research develops the EXAONE model family, and it finished first in the opening stage of the government’s sovereign model project in January 2026. In August, it advanced again as one of the last three teams with K-EXAONE, which LG says it scaled to more than three times its earlier size.

LG is also widening access. On 9 November 2026, it opened the K-EXAONE 2.5 API free of charge to developers, start-ups and researchers, reporting a gain of more than 70% across 24 benchmarks against its July release, based on its own tests. It’s aiming for a frontier-level EXAONE model by January 2027. Second-round evaluators praised its agentic AI and trustworthy-AI work.

Key Solutions:

• ✔ K-EXAONE sovereign foundation models
• ✔ Free K-EXAONE 2.5 API for developers
• ✔ EXAONE deployments for electronics, finance, telecoms and biotech

Best for: Manufacturers, research teams and developers who want a top-ranked Korean model with open access.

3. SK Telecom

SK Telecom, the SK Group’s mobile operator, traces its roots to 1984 and is based in Seoul. In January 2026, its team unveiled A.X K1, a 519-billion-parameter model it called the first Korean model above 500 billion parameters. It’s released under the Apache 2.0 licence, which permits commercial use, and it tied with LG for first place in the project’s first-phase benchmark.

The company’s edge is scale. Its consortium spans more than 20 organisations, including SK hynix and KAIST, and it made the final three in August 2026 with A.X K2, a 688-billion-parameter successor. Evaluators rated its maths reasoning and Korean-language performance among the best of the models compared, and noted it already runs inside a large commercial service. A.X K1 runs on Rebellions’ RebelServer, so SK Telecom can offer a model-to-chip stack built largely at home.

Key Solutions:

• ✔ A.X K1 open-weight large language model
• ✔ Multimodal upgrades covering images, voice and video
• ✔ AI for telecoms and defence

Best for: Large organisations that want a very large open model to adapt and host themselves.

4. Samsung SDS

Founded in 1985 as Samsung Data Systems and listed on the Korea Exchange, Samsung SDS is led by CEO Lee Jun-hee from Seoul. It sells a full enterprise AI stack: FabriX, a generative AI platform that links language models to company systems; Brity Copilot and Brity Meeting for daily work; and Samsung Cloud Platform underneath.

It has real pull with large buyers. Samsung SDS was the first Korean reseller partner for ChatGPT Enterprise and works on a government-wide common AI infrastructure project with the Ministry of the Interior and Safety. In November 2026, it began building a 60-megawatt AI data centre in Gumi, backed by a 427.3 billion won investment and designed to expand to 120 megawatts. In July, it launched a cloud service on FuriosaAI chips.

Key Solutions:

• ✔ FabriX generative AI platform
• ✔ Brity Copilot, Brity Works and Brity Meeting
• ✔ Personal AI agents for email, meetings and briefings
• ✔ ChatGPT Enterprise resale

Best for: Large companies and public bodies rolling out AI to thousands of staff.

5. Upstage

Founded in 2020 by Kim Sung-hoon, who previously led Naver’s Clova team, Upstage builds the Solar family of large language models. In May 2026, it completed the acquisition of AXZ, operator of the Daum portal, from Kakao through a share swap that left Kakao with a minority stake. Upstage plans to pair Solar with Daum’s search engine and data for a context-aware AI search service.

It’s also one of the three teams left in the sovereign model project, with evaluators citing its full-stack approach, Daum integration and use of a domestic NPU. Revenue rose from 5.9 billion won in 2022 to 24.8 billion won in 2025, and clients include Shinhan Investment and K Bank. Amazon and AMD have invested, and KB Securities and Mirae Asset Securities are lead underwriters for a planned IPO.

Key Solutions:

• ✔ Solar large language models, available on AWS
• ✔ Context-based AI search through Daum
• ✔ Document AI for finance, legal and public-sector work

Best for: Banks, insurers and law firms that need strong Korean document processing.

6. FuriosaAI

Seoul-based FuriosaAI was founded in 2017 by June Paik, a former Samsung Electronics engineer who remains CEO. It designs chips for AI inference, the work of running trained models. Its second-generation chip, RNGD, entered mass production in January 2026, with TSMC producing 20,000 units this year and more than 40,000 planned for 2027.

Investors have noticed. On 28 October 2026, FuriosaAI announced a $620 million Series D at a valuation of about $2.9 billion, with South Korea’s National Growth Fund putting in $280 million and Naver among existing shareholders. In 2025, it turned down an $800 million acquisition offer from Meta and signed LG AI Research as a customer. Samsung SDS and Upstage now use RNGD too.

Key Solutions:

• ✔ RNGD inference accelerator, rated at 180 watts with air cooling
• ✔ Inference hardware tuned for LLMs such as EXAONE
• ✔ Cloud and on-premises deployment

Best for: Data-centre and cloud operators looking to cut the power cost of running large models.

7. Rebellions

Rebellions was founded in 2020 and is led by CEO Park Sung-hyun from Seongnam, just south of Seoul. In December 2024, it completed a merger with Sapeon Korea, SK Telecom’s AI chip unit, at a combined value of about 1.3 trillion won. The merged company kept the Rebellions name, with Park as sole CEO.

It now sells complete inference systems rather than chips alone: RebelServer, RebelPOD and RebelRack, built on Rebel100 silicon. In March 2026, it raised $400 million in a pre-IPO round at about $2.34 billion, led by Mirae Asset Financial Group and the Korea National Growth Fund, taking total funding to $850 million. It has set up entities in the US, Japan, Saudi Arabia and Taiwan, and said in July 2026 that it aims to list in Korea in the first half of 2027.

Key Solutions:

• ✔ RebelPOD turnkey inference infrastructure
• ✔ RebelRack for generative and agentic AI workloads
• ✔ RebelServer, which runs SK Telecom’s A.X K1

Best for: Telecom operators, governments and neoclouds building inference capacity.

8. Kakao

Kakao, the Jeju-headquartered group behind KakaoTalk, Korea’s most widely used messaging platform, puts AI where its users already are. In February 2025, it became the first Korean company to sign a strategic partnership with OpenAI, and it launched ChatGPT for Kakao in October 2025. By May 2026, the service had passed 11 million cumulative subscribers.

Under CEO Chung Shin-a, Kakao follows an AI model orchestration strategy, combining its own Kanana models with external ones such as OpenAI’s. In June 2026, it added a ChatGPT chatbot that users can call up inside group and one-on-one chatrooms. Kakao wasn’t selected for the sovereign model project and sold Daum to Upstage this year, but few rivals can match its consumer reach.

Key Solutions:

• ✔ ChatGPT for Kakao inside KakaoTalk
• ✔ Kanana AI agent and in-house models
• ✔ ChatGPT chatbot for KakaoTalk chatrooms

Best for: Consumer brands and services that want AI-powered reach among Korean users.

9. Lunit

Lunit started in 2013 and is based in Gangnam, Seoul. The KOSDAQ-listed company builds AI for cancer screening and precision oncology. Lunit INSIGHT products read mammograms, 3D breast scans and chest X-rays, while Lunit SCOPE analyses tissue to help guide treatment. It has also completed the acquisition of Volpara, a breast-imaging software company.

Few artificial intelligence companies in Korea have this kind of overseas traction. More than 10,000 medical institutions worldwide have adopted its AI, the company says. In the first half of 2026, revenue hit a record 45.8 billion won, up 23% year on year, with about 95% from overseas, while its operating loss narrowed by around 31%. Seo Beom-seok is CEO.

Key Solutions:

• ✔ Lunit INSIGHT MMG and DBT for breast cancer screening
• ✔ Lunit INSIGHT CXR for chest X-rays
• ✔ Lunit SCOPE AI biomarker and pathology tools

Best for: Hospitals, national screening programmes and pharma companies working on cancer.

10. TwelveLabs

TwelveLabs was founded in 2021 by Jae Lee and friends he met while serving in the Republic of Korea’s Cyber Operations Command. It’s headquartered in San Francisco, keeps offices in Seoul and Pangyo, and builds models that understand video. Marengo turns footage into searchable data, while Pegasus converts video into text and time-stamped metadata.

It has strong backing on both sides of the Pacific. In July 2026, it raised a $100 million Series B co-led by NEA and Naver Ventures, with Amazon taking part. Its models are available through its own API and Amazon Bedrock, and it has a multi-year deal to run video inference on AWS Trainium chips. Korean customers include broadcaster SBS and GS SHOP.

Key Solutions:

• ✔ Marengo video embedding model
• ✔ Pegasus video language model
• ✔ Search, Analyze and Embed APIs, including via Amazon Bedrock

Best for: Broadcasters, sports bodies, advertisers and owners of large video archives.

How to Choose the Right AI Partner in South Korea

Decide which layer you need

Korean AI companies sell very different things: models to build on, finished platforms for staff, or hardware for your data centre. Know which layer you’re buying first.

Check where data lives

If you handle financial, health or government data, ask exactly where models are trained and hosted. Public cloud, private cloud and on-premises options carry different risks.

Test on your own documents

Many benchmarks are vendor-reported. Run a pilot on your own contracts, reports or customer chats, in Korean and any other languages you use, before you commit.

Read the licence

Open models such as A.X K1 use Apache 2.0, while others have custom licences or paid APIs. Check the terms for commercial use and fine-tuning.

Weigh financial stability and hardware

Many AI startups in South Korea are still loss-making and preparing for IPOs, so check a vendor’s runway. Ask too whether it runs on Nvidia GPUs or Korean NPUs, since that affects cost as you scale. If you’re also weighing global tools, our guide to the top 10 generative AI platforms is a useful reference.

FAQs

Which are the biggest AI companies in South Korea?

By scale, the leaders are Naver, LG AI Research, SK Telecom, Samsung SDS and Kakao, each combining models or platforms with large data centres or user bases. Upstage leads the independent model developers.

What is Korea’s sovereign AI foundation model project?

It’s a Ministry of Science and ICT programme that funds rival teams to build a Korean foundation model and narrows the field in stages. After the August 2026 evaluation, LG AI Research, SK Telecom and Upstage remained, and the next stage will narrow the field further.

Which Korean AI chip companies should I know?

FuriosaAI and Rebellions lead in inference. FuriosaAI makes the RNGD chip, while Rebellions sells its silicon inside complete server and rack systems.

Conclusion

The best Korean AI companies now cover the full stack, from silicon to models to apps used every day. Conglomerates bring scale, while specialists such as FuriosaAI, Lunit and TwelveLabs have shown they can win customers far beyond Korea.

The field will keep shifting as the sovereign contest narrows and IPOs arrive. Pick a partner by the layer you need, test it on your own data, and watch who survives the next round.