South Korea builds free national AI chatbot to cut US dependence

By Mei Lin Chen · 2026-07-28

Seoul moves first

South Korea's Ministry of Science and ICT opened the AI for All Project competition on 13 July 2026, giving bidders until 11 August to pitch a general-purpose AI chatbot that would be free of charge and unlimited for every citizen before year's end.verifiedVerified Source: msit.go.kr^[1] The rules are unambiguous: selected providers must draw at least 50 per cent of their service from domestically developed AI models, up to 30 per cent from other Korean companies, with foreign models permitted only at the margins.^[1]

Winners receive 512 Nvidia B200 GPUs to support an early launch.^[1] National budget funding to cover the full cost of citizen access follows from 2027. MSIT also announced a public AI agent alongside the chatbot, framing both as foundational infrastructure rather than a consumer product.

Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon put the vision plainly. "AI for All is more than a simple service," Bae said. "It is the calculator and computer of the AI era, enabling people to work, learn, and go about their daily lives with AI. It is also a platform that ensures everyone can share fairly in the benefits of the new economic structure that AI will bring about."^[1]

The sovereignty rationale

Three days after the tender opened, MSIT briefed President Lee Jae Myung on the broader objective: building an "AI Universal Basic Society" anchored to a sovereign AI model.^[2] The ministry named the direct trigger without hedging: as the United States tightens AI-related access controls, securing a homegrown model has become a national-security and data-sovereignty priority.verifiedVerified Source: msit.go.kr^[2]

South Korea is not building a domestic chatbot because it can. Seoul has concluded it cannot afford not to. The explicit linkage between US export controls and South Korea's AI policy signals a strategic calculation that other US-aligned nations will face in their own time.

A global pattern takes shape

On 21 July 2026, Microsoft announced a multibillion-dollar commitment to deploy frontier AI through Mistral's Europe-based GPU infrastructure, extending sovereign cloud offerings across Foundry, Copilot Studio and Azure to let regulated European industries keep data out of US-jurisdiction systems.verifiedVerified Source: news.microsoft.com^[3] Microsoft Vice Chair and President Brad Smith put the case directly: "Europe should have access to the world's most capable AI without compromising control over their data, operations or digital future."^[3]

India's BharatGPT initiative, developed under the Digital Public Infrastructure programme, follows the same logic. Japan is exploring national AI legislative frameworks aimed at equivalent strategic autonomy. Governments previously content to licence American AI are now treating domestic model capability as a strategic asset, not a convenience.

What Australia has done

The Australian Government established the Office of AI inside the Department of the Prime Minister and Cabinet on 15 July 2026, tasked with coordinating national AI standards and unlocking AI adoption across the public service.^[4] The infrastructure supporting that office is, for now, foreign. The Digital Transformation Agency secured a five-year Volume Sourcing Agreement with Microsoft on 26 February 2026, locking in stable pricing, capped price increases and enhanced data protections for ICT products and services across the Australian Government.^[5]

On 1 April 2026, Australia signed a collaborative Memorandum of Understanding with Anthropic covering AI safety, technical exchanges and emerging risk analysis through the AI Safety Institute.^[6] Minister for Industry and Innovation Tim Ayres said the agreement showed Australia was "open for business, where investment aligns with Australia's priorities and Australian values," adding that the government and Anthropic were working together to "harness AI responsibly."^[6]

The gap that remains

Australia holds a coordination office, a Microsoft licensing deal and a safety-focused MoU. What it does not hold is a domestic AI model, a funded pathway to one, or a requirement that public-sector AI services use Australian-built technology. Every query a citizen or public servant routes through Copilot or Claude traverses infrastructure owned and operated by foreign companies subject to foreign law.

South Korea's MSIT has said explicitly that this arrangement is a vulnerability. The European Union reached the same conclusion and structured the Microsoft-Mistral deal accordingly. Australia has not made that call, and the Office of AI's mandate, as published on 15 July 2026, does not include developing or procuring a sovereign model.^[4]

What comes next

South Korea's tender closes on 11 August 2026, after which MSIT will select providers and begin deploying the 512 B200 GPUs.^[1] The citizen-facing service is due before the end of 2026, with national budget coverage of ongoing costs beginning in 2027. Australia's Office of AI, established four days before South Korea opened its tender, has published no equivalent procurement timeline.

Mei Lin Chen

Mei Lin Chen writes about trade, supply chains and manufacturing across Asia and Australia. She follows goods from the factory floor to the shelf.