·
Updated 2026-07-06
This is Part 3 of “Korea’s AI Bet,” a five-part series on how Korea is trying to build an AI economy — and what it means for foreigners. Part 1 argued Korea treats AI as infrastructure, not a tool. Part 2 asked who pays for “AI for All”. This is Korea’s middle-power playbook in action: how the rest of the world — superpowers, middle powers, and the multilateral bodies they all sit inside — is responding to both, and a pattern that keeps repeating: Korea keeps getting invited to the table by governments and companies who otherwise agree on very little.
The Rush Nobody Quite Announced
Parts 1 and 2 of this series were, in a sense, a domestic story — what Korea is building for its own citizens and who foots the bill. This part is the same bet turned outward: once a government commits real money and law to an AI project at home, it needs the world’s frontier labs, chip suppliers, and rule-making bodies to cooperate rather than box it in. That’s a different kind of diplomacy than a trade talk or a defense pact, and it explains why a single ministry-level unit now exists just to run it.
Line up the last year and a half and the pattern looks less like coincidence than a deliberate position. In February 2025, Korea’s Ministry of Foreign Affairs created a dedicated International AI Diplomacy Division — a standing unit whose entire mandate is managing AI relationships with other governments. Since then the agreements haven’t stopped: an MOU with OpenAI on regional infrastructure and public-sector adoption (Oct 1, 2025); a “Technology Prosperity Deal” with the US at the APEC summit in Gyeongju (Oct 29); a digital-cooperation MOU with Singapore and an innovation-startup MOU with China within the same week (Nov 1–2); an AI-industry MOU with Indonesia (Apr 1, 2026); a Strategic AI Cooperation Framework with the UAE (May 13); an MOU with Saudi Arabia expanding energy ties into AI (mid-June); and a partnership with Google DeepMind that brought the company’s first AI campus outside its home country to Seoul (Apr 27). That’s roughly one notable AI cooperation agreement a month for over a year — the frequency is the story here, not any single deal.
Korea isn’t new to convening, either. Two years before any of this, in May 2024, Korea co-chaired the AI Seoul Summit with the UK, producing the Seoul Declaration on safe, innovative, and inclusive AI — the first time a country other than the US or UK had run one of these global AI safety summits. The current deal rush reads less like a sudden discovery than Korea cashing in a credential it started building two years earlier.
Why Everyone Wants a Meeting With Seoul
Korea’s leverage here isn’t goodwill, it’s structural. Samsung and SK Hynix together control an estimated 90% of the world’s high-bandwidth memory (HBM) market — the chip technology every frontier AI training cluster depends on. That gives Korea a genuinely rare position: an economy both the US and China need for their AI supply chains, with a security alliance in Washington and its largest trading relationship in Beijing. Analysts have started naming this directly — a January 2026 East Asia Forum piece called it a “forced US-China balancing act”; an April 2026 Carnegie Endowment paper called it “governing AI in the shadow of giants.” Academics studying AI governance have started naming this position directly, too: a recent paper on the global politics of AI standard-setting groups Korea with Singapore and the UAE as “swing states” — economies with enough chips, capital, or compute to matter to both blocs, using strategic ambiguity rather than alignment as their actual leverage. (Worth being precise about what this isn’t: a separate February 2026 Carnegie Endowment report on “South-South AI collaboration” describes something different — developing countries like India and African nations partnering directly with each other. Korea, an OECD member and aid donor, isn’t a Global South economy in that sense; “swing state,” not “South-South,” is the more accurate label for what Korea is doing.) That framing isn’t only coming from Washington-adjacent think tanks and journals: at the Kyunghyang Forum in Seoul in June 2026, Indonesia’s presidential trade envoy Mari Elka Pangestu made a related argument from the other side of the table, noting that the US and its allies control roughly 65% of the world’s AI compute capacity and China another 20% — and naming Korea specifically as a candidate to lead a “middle-power coalition” alongside ASEAN’s 680 million consumers. That’s a developing-country diplomat naming Korea as a partner, not Korea marketing itself as one.
The chip angle isn’t static, either. Korea sits inside the informal “Chip 4” grouping with the US, Japan, and Taiwan — four economies controlling an estimated 82% of global semiconductor market share and up to 99% of the memory-chip supply chain. Washington keeps tightening export controls around exactly that exposure: a January 2026 Commerce Department rule moved licensing for advanced AI chips like Nvidia’s H200 and AMD’s MI325X from near-automatic denial to case-by-case review. Korea’s two largest chipmakers sit on both sides of that line — bound by US rules on advanced equipment, but still running China-exposed fabs and joint ventures neither wants to abandon. That’s the real tightrope: not an abstract ethics stance, but a concrete question of which government approves which shipment. It’s a domestic political question too, not just an external analyst’s framing — closer defense-and-AI cooperation with Washington has already drawn commentary inside Korea that it complicates ties with Beijing and tests President Lee Jae-myung’s own label for his foreign policy, “pragmatic diplomacy.”
Put together, the logic isn’t “pick the US AI stack or the Chinese one” — it’s “stay useful to both, and to everyone neither one prioritizes.” That splits into four tracks, each moving on its own calendar but pointing the same direction.
Track One: Government to Government
The clearest evidence is the density of state-to-state deals signed in under a year.
| Partner | What was signed | When |
|---|---|---|
| United States | “Technology Prosperity Deal” — cooperation across AI hardware, models, software, and standards | Oct 29, 2025 (APEC Gyeongju) |
| Singapore | Digital-cooperation MOU, plus seven company-to-company MOUs on autonomous driving and public safety | Nov 2, 2025 |
| China | Innovation-startup MOU spanning AI, biopharma, and green industry — joint investment, standards, talent exchange | Nov 1, 2025 |
| Philippines | IP cooperation covering AI use in patent examination | 2025–26 |
| APEC (multilateral) | AI Initiative (2026–2030); Korea to host a new Asia-Pacific AI Center | Gyeongju Leaders’ Declaration, 2025 |
| ASEAN (multilateral) | Hanoi Digital Declaration — AI cooperation, digital infrastructure, workforce, trusted data flows | Jan 15–16, 2026 |
| Indonesia | AI-industry MOU, including a joint “Global AI Universal Basic Society Solidarity Initiative” | Apr 1, 2026 |
| India | “India-Korea Digital Bridge” — AI, semiconductors, electronics/IT | Apr 2026 |
| UAE | Strategic AI Cooperation Framework; Korea joins a UAE AI-and-energy megaproject reported at roughly ₩30 trillion | May 13, 2026 |
| Saudi Arabia | MOU expanding energy ties to cover AI, critical minerals, digital transformation | Jun 13–14, 2026 |
Qatar’s trade minister also brought a delegation to Seoul the same season for AI and semiconductor talks — no signed deal yet, but one more Gulf capital in the queue.
Two things stand out. The US and China deals landed within a month of each other — Korea didn’t hide either signing to avoid the optics of playing both sides. And four of the partners above — India, Indonesia, the UAE, and Saudi Arabia — aren’t superpowers at all; they’re other middle powers running a similar hedge, which is different evidence than “the US and China both want Seoul.” It suggests the positioning is legible to other countries doing the same calculation, not just useful to the two governments Korea is trying not to choose between.
Two of these deals are worth unpacking further, because the specifics show more than a courtesy signature. The UAE framework grew out of a joint working group set up during a state visit in November 2025, and by May it had produced concrete projects, not just language: Korea joining the UAE’s own “Stargate” AI-and-energy buildout — a 200-megawatt data-center campus in Abu Dhabi scaling toward 5 gigawatts, powered by a nuclear-gas-renewables mix — plus a separate “AI Port Logistics” project applying physical AI and robotics to port operations in both Busan and Abu Dhabi’s Khalifa Port. The India deal is just as concrete about the division of labor: the “Digital Bridge,” unveiled during President Lee Jae-myung’s state visit to India (the first by a Korean head of state in eight years), pairs India’s AI software and engineering talent with Korea’s semiconductor fabrication and precision-manufacturing base, with both governments setting a $50 billion bilateral trade target by 2030. Neither deal is a press-conference photo op with no follow-through — both specify what each side actually supplies, and by when.
Hosting the APEC AI center matters more than the ceremonial framing suggests: it sets the agenda for the data standards and safety benchmarks the whole region works from for five years, a role only offered to a country other members trust as a non-partisan convener. Layered against the Singapore, Indonesia, and ASEAN-wide deals — all aimed at Southeast Asia — the pattern looks like Korea building a standards hub for a region neither superpower fully controls, rather than trying to out-invest either one directly.
Track Two: Government to Frontier Labs
The second track is about who inside the AI industry the government deals with directly. The OpenAI MOU covers regional infrastructure, public-sector adoption, and Korean-company participation in Stargate — OpenAI’s roughly $500 billion data-center buildout with Oracle and SoftBank. Six months later, Korea’s deputy prime minister signed a separate MOU with Google DeepMind covering joint research, alongside confirmation that Seoul would host Google’s first AI campus outside its home country. In between, the government has also been in talks with Anthropic, described in Korean coverage as diversifying “away from an OpenAI-only” posture. None of this reads as betting on a single lab — it reads as a government methodically signing with competitors because it wants access to all of them, not an alliance with any one.
Securing a role inside Stargate — rather than buying finished services from it — is a materially different deal than a vendor contract; that kind of infrastructure-level access usually goes to a small handful of trusted partners. And choosing Seoul over larger hubs like Tokyo or Singapore for Google’s first campus outside its home country is itself a signal: DeepMind picks locations based on research access, not market size.
The obvious objection is worth naming directly: every company here — OpenAI, Google, Anthropic — is American. If “won’t pick a side” rested on this track alone, that would be dependency with a diplomatic layer on top, not neutrality. An April 2026 Nikkei Asia piece on “AI colonialism” makes exactly this case — foreign AI dependence risks capital outflow, data exposure, and interference from whichever government the developer ultimately answers to. That risk is real, not imaginary. But this track isn’t what carries the “non-aligned” argument here — Track One’s China MOU and Track Four’s open-source release do that work instead. Track Two looks entirely American because the world’s frontier labs currently are, not because Korea passed over an equivalent Chinese lab that wanted the same physical campus and government MOU. No such lab has made that ask yet.
Track Three: Setting the Rules, Not Just Signing Deals
The first two tracks are bilateral — one government or company at a time. The third is different in kind: Korea helping write the rules that outlast any single deal. Korea was one of fifteen founding members of the Global Partnership on Artificial Intelligence (GPAI) when it launched in 2020, alongside the US, EU, Japan, India, and the UK. GPAI merged into the OECD in 2024, and in 2026 Korea moved from founding member to co-chair — the same deputy prime minister who signed the DeepMind MOU and frames Korea’s sovereign-AI project as “inclusive, not closed” now co-chairs GPAI alongside Singapore’s digital minister. Ministers met in New Delhi in February 2026, and GPAI welcomed Malta and Saudi Arabia as new members, growing to 44 countries across six continents.
That February meeting wasn’t a ceremonial photo-op either — 58 delegations attended, 20 of them led by a minister or above, and the agenda included building a shared AI Policy Toolkit for countries outside GPAI’s own membership and examining what AI actually does to agriculture, a sector middle-income economies care about as much as chip fabs. GPAI also runs three standing expert centers, in Paris, Montreal, and Tokyo, that feed technical research into the policy conversations ministers have at moments like this one.
One official running the frontier-lab MOU and the open-source framing while also co-chairing the body writing global AI governance norms is a small sign this isn’t three ministries acting independently. Add the APEC AI center Korea is set to host and the 2024 Seoul Declaration that preceded all of it, and the picture is clearer: Korea isn’t only collecting bilateral MOUs, it’s building a seat at every table where AI governance actually gets negotiated.
That seat is worth being honest about, too. GPAI and the APEC initiative are consensus-based and non-binding — a co-chair role buys agenda-setting influence and credibility, not the power to force China, the US, or anyone else to adopt Korean-preferred standards. This track is a slow bet: that being in the room when standards get drafted matters more over five or ten years than any single MOU signed this year, and one Korea can keep making regardless of who wins the next election in Washington, Beijing, or Seoul. That combination of agenda-setting patience and specific deliverables — the Policy Toolkit, the agriculture study — is what separates a seat at the table from a chair with a country’s name on it.
Track Four: The Open-Source Bet
The clearest test of whether Korea’s openness is more than rhetoric comes in August 2026, when the government plans to release its sovereign foundation model as fully open source. Domestically it’s known as 독파모 — short for “독자 AI 파운데이션 모델” (“independent AI foundation model”), also nicknamed the “national-team AI” project. Five companies — the same shortlist covered in Part 2: LG AI Research, SK Telecom, Upstage, and Motif — cleared the program’s first-stage evaluation against global benchmarks in early 2026, and the stated plan is to publish the results on platforms like Hugging Face with no country restrictions — available to a developer in Seoul or Lagos on the same terms. That’s not an inference from the deployment plan; it’s how Korea’s deputy prime minister, who oversees the program, describes it himself: “I’m not someone who argued for closed sovereign AI. Inclusive AI matters more,” framing the goal as a model “the world can choose” rather than one built purely for domestic control. The same inclusive language shows up bilaterally — Korea’s April 2026 MOU with Indonesia created a joint “Global AI Universal Basic Society Solidarity Initiative” using nearly identical framing. The August release reads as the technical follow-through on a promise Korea had already started making one country at a time.
That framing is being tested at home, too, and not everyone is convinced. Korean AI developers including Upstage and Naver reportedly diverge on what “sovereign” should even mean here, and some industry critics go further: if a domestically-branded model shares an encoder or core weights with an existing foreign model, calling it “sovereign” overstates its independence. That’s close to the same dependency question Nikkei raised about Track Two, aimed inward instead of outward — a reminder that “open source, built in Korea” and “built independently by Korea” are different claims, and this piece is only making the first one.
An open, unaligned Korean model doesn’t resolve the split between a US-aligned and a China-aligned AI stack, but it gives countries that don’t want to choose a third option. It won’t match GPT-5-class or Gemini-class raw capability, and no one serious claims it will — but for a mid-tier model, it removes the licensing and alignment questions entirely, a different offer than either superpower’s frontier labs currently extend to non-aligned countries. For the practical version of this — what “AI for All” gives you access to right now — Part 4 of this series covers the free AI tools available in Korea.
The Gap Between the Pitch and the Delivery
None of this means Korea’s openness has already reached the developing-world audience its own messaging targets. The deepest, best-funded deals are still overwhelmingly with wealthy, AI-capable partners — the US, Singapore, China, the UAE, Saudi Arabia, and three of the best-funded labs on earth. Some of that gap has closed faster than expected: India’s Digital Bridge and Indonesia’s MOU are genuine middle-power and developing-country partnerships, not wealthy-core deals, and arrived within weeks of the Gulf agreements. The clearest remaining gap is Africa — a KOAFEC ministerial conference scheduled for Seoul in September 2026 is explicitly themed around AI and digital infrastructure for Africa, and Korea has separately agreed with Italy to explore joint AI and semiconductor projects in African countries including Uganda, Ethiopia, Egypt, and Côte d’Ivoire. Those are real institutional channels, not nothing — but neither is yet a signed, bilateral Korea-Africa AI MOU at the density of the Gulf or Southeast Asian deals above.
This is a familiar pattern from prior technology cycles: telecom, vaccines, and renewable energy all saw wealthy-country “open access” pledges reach low-income countries years later than promised, usually because the first wave goes to partners who can already absorb the technology. There’s no evidence yet that Korea’s rollout avoids that pattern entirely — the fairer read is that it’s more mixed than either the optimistic or skeptical version suggests: genuinely broader than “wealthy core only” thanks to India and Indonesia, not yet broad enough to include Africa or the lowest-income parts of Southeast Asia. Three markers would separate delivered follow-through from stated intent: a signed bilateral AI agreement with an African government; verifiable download data for 독파모 from developing-country IP ranges once it ships in August; and a funded capacity-building program under APEC’s AI center or GPAI that reaches beyond their wealthier core members. None exist yet in confirmed form; all three are plausible within the next 12–18 months given the diplomatic infrastructure already built. Whether they materialize is the most useful thing to track about this story going forward.
Stay Updated
Diplomatic momentum can reverse faster than visa quotas do — a change in government in Washington, Beijing, or Seoul could shift any of these relationships within a single term. We’re tracking this as a trend, not a fixed state, and will update the numbers as new deals land or old ones stall. If any of this changes your own calculus about founding or hiring in Korea, our F-2-7 Points Calculator, D-8 founder visa guide, and EOR vs. own-entity guide cover the individual and company paths.
Disclaimer: This post reflects the author’s experience and publicly available information as of 2026. It is general information, not legal, tax, or immigration advice. Rules and rates change — verify current details with the relevant authority (NPS, NTS, MOJ) or a licensed professional before acting.
