Google DeepMind Picks Seoul for Its First Campus Outside London

Gangnam

DeepMind is opening its first campus outside its London King's Cross headquarters. The location: Seoul's Gangnam district. Opening within 2026, with an initial 50-person hiring plan that mixes researchers, engineers, and language specialists. It's a research campus — not a sales office. That distinction matters.

DeepMind started in London in 2010 and stayed in London after Google acquired it in 2014. Sixteen years of research output — AlphaGo, AlphaFold, Gemini — all from one building. This is the first crack in that single-HQ posture, and it's going to Seoul. Not Tokyo. Not Beijing. Not Singapore.

The reasoning is concrete. Korea is (1) the home of Samsung's and SK Hynix's memory and chip supply chain, (2) one of the top-four global AI talent pools, (3) a vertical-industry leader in mobile, gaming, and content. Add Korea's same-week announcement of a $43B (60 trillion won) strategic-technology investment package, and the timing aligns.

Why each side is moving

For DeepMind, two signals. Language depth: Korean's morphological complexity makes LLM tokenization less efficient than English. Closing that gap inside Korea — with Korean-native engineers and linguists — is faster than doing it remotely. Local industrial data access: Samsung collaboration scenarios benefit from physical proximity to domain experts and data partners.

For Google, the larger context is Sovereign AI. Governments globally want AI systems trained, hosted, and aligned with their data, culture, and regulation. A first campus outside US/UK is the move that operationalizes that pitch. Hassabis's framing — "Korea's combination of talent and industry density is rare globally" — is the public version of the internal decision memo.

For the Korean government, the announcement is the first visible payoff of the "AI 3rd power after US and China" policy line. Minister Yoo Sang-im flagged "global AI HQ attraction" as a KPI at last year's National Digital Conference.

For Samsung, the alignment is collaborative rather than competitive. Samsung Research's in-house Gauss LLM doesn't compete head-on with DeepMind. Memory-chip simulation, on-device AI, and other adjacent areas can yield mutual gains.

What's announced

Item	Seoul campus	London HQ (reference)	Prior baseline
Location	Seoul, Gangnam-gu	London King's Cross	—
Year-1 hiring	50 (2026)	1,500+	0
2-3 year goal	200-300	2,500+ (est. 2030)	0
Core teams	Korean LLM, on-device AI, industry partnerships	Core models, safety, applied	—
5-year investment	~₩1T (~$725M) est.	undisclosed	0
Government partner	MSIT + Seoul Metro	UK AI Safety Institute	—

The ₩1T investment is roughly 1.7% of Korea's ₩60T strategic tech package. As a single FDI move into Korean AI, it ranks among the largest ever. By comparison, NVIDIA's Korea R&D announcement last year was ~₩500B and Microsoft Korea's AI investment ~₩700B — DeepMind exceeds both combined.

Hiring is mostly Korean talent, with roughly 70-80% expected local. The remainder is APAC talent transferring to Seoul. Bilingual (English + Korean) operations from day one.

Who wins what

DeepMind. Faster Korean-language model improvement, Korean industry data access, Gemini APAC market acceleration. Claude has been strong in Japan and Korea — Seoul campus is the direct counter.

Google. The Sovereign AI playbook gets a reference customer. Negotiations with India, Japan, UAE, Saudi Arabia have reportedly started; "we did this in Korea" is the case study.

Korean government. Policy KPI achievement. FDI statistics in AI jump by a trillion-won bracket. Sets the template for further attempted attractions of OpenAI and Anthropic.

Samsung. First-mover access to DeepMind collaboration on memory simulation and on-device optimization. Complementary, not competitive, to Gauss.

Korean AI talent. New employer at the top of the global ladder, inside the country. Compensation expected at ~1.5-2× the local market for senior roles, raising broader local salary pressure.

Historical comparisons

Microsoft Korea AI Lab (2024, ~₩700B). Started with Azure OpenAI Korea region + Korean LLM team. Grew to ~200 staff in 18 months. Worked.

NVIDIA Korea R&D (2025, ~₩500B). CUDA documentation Korean, autonomous-driving simulation, gaming/media collaboration. Synergized with Korean game studios.

Tesla Korea AI (2023, partial failure). FSD Korean localization. Stalled within a year due to road-data access and regulatory friction. Lesson: foreign AI HQs in Korea aren't automatic wins.

DeepMind enters as a research campus, with less data/regulatory friction than Tesla had. Tesla pattern unlikely to repeat.

How rivals counter

OpenAI. Korean expansion rumored late 2025. May accelerate; Microsoft Korea synergy is the natural shape rather than a standalone office. Anthropic. Smaller scale than DeepMind. Likely picks Japan or Korea — not both — for its first APAC campus. Chinese AI. Direct entry blocked politically. Indirect via B2B with Korean firms. Korean LLMs (Naver HyperCLOVA, KT, Kakao, Samsung Gauss). DeepMind raises talent competition but also lifts overall market interest. Government may funnel more of the ₩60T to local champions to balance.

What this changes for you

Engineers. Top-tier global lab now hires inside Korea. Senior ML, Korean NLP, GPU infra roles in highest demand. Founders. New collaboration ecosystem will form around Gangnam. AI tools, evaluation, agent startups gain a market. Investors. Korean AI valuations get a re-rate. Korean LLM next-round prices most affected. General users. Korean-language Gemini quality should improve within 12 months. Re-test ChatGPT vs. Gemini vs. Claude in Korean every six months.

Stakes

• Wins: Korean government (KPI met), Korean AI talent (top-tier hiring locally), Google DeepMind (Korean market + Sovereign AI template), Samsung (collaboration first-mover).
• Loses: Anthropic (pressure to enter APAC physically), Korean LLM labs (talent competition).
• Watching: OpenAI (will entry timing accelerate?), Korean government (₩60T allocation between local LLMs vs. foreign attraction infrastructure).

Skeptics, named

Park Chan-ik (Yonsei IT Graduate School, AI policy researcher) wrote in Hankyoreh that "foreign AI HQ attraction can be a different shape of brain drain." Korean talent stays in Korea, but IP, data, and models flow to HQ. He argues the deal needs IP-sharing or domestic-deployment-priority clauses to count as genuine sovereignty.

Lee Kyung-jeon (Kyung Hee Business School) questioned whether the ₩60T allocation balances local LLM funding versus foreign-attraction infrastructure. The next year of policy debate will be about that split.

Internal talent flow is the third concern. Senior engineers at HyperCLOVA, KT LLM, and Samsung Gauss may move to DeepMind; the system-level effect on the Korean AI ecosystem will play out over 12-24 months.

Tomorrow morning

Engineers: Set an alert on DeepMind careers page for Seoul roles. Polish your Korean-NLP portfolio if you have one. Founders: Track Gangnam-area AI ecosystem formation. DeepMind's partner-call program (likely H2 2026) is the watch item. Investors: Monitor Korean AI fund round prices every six months. Watch for re-rate from DeepMind effect. Users: Score Korean-language Gemini, GPT, Claude on identical prompts now. Re-score in six months to measure DeepMind impact.

Sources

• EconMingle — DeepMind Seoul: https://econmingle.com/economy/google-deepmind-seoul-ai-campus-2026/
• DeepMind: https://deepmind.google/about/
• Korea MSIT ₩60T announcement: https://www.msit.go.kr/
• ZDNet Korea — AI infra policy: https://zdnet.co.kr/
• Reuters — Google APAC investment: https://www.reuters.com/technology/