South Korea is doubling down on AI. On Monday, November 2, 2026, Reuters reported that President Lee Jae Myung vowed about $15.97 billion for AI, chips, physical AI and data centers in the country's 2027 budget, nearly twice the 2026 amount. This post separates what that headline number covers, what the government has already put in writing, and what is still a proposal.
One caveat first: we could not open the full Reuters article, so the Monday remarks are attributed to Reuters as summarized in its headline. The detailed figures below come from the Korean government's September budget briefing as reported by The Korea Times, and from July coverage of Lee's fiscal strategy meeting.
TL;DR: the Korea AI budget at a glance
| Item | What is reported | Source |
|---|---|---|
| Headline pledge | About $15.97B for AI, chips, physical AI, data centers; nearly 2x 2026 | Reuters (headline) |
| Three megaprojects | Semiconductors, physical AI, AI data centers | The Korea Times |
| Funding for them | 21.3 trillion won, up 97.2% from this year | The Korea Times |
| Physical AI demos | 2.6 trillion won | The Korea Times |
| National model support | 10,000 GPUs plus data and workers | The Korea Times |
| Chip policy | A special account for semiconductors | The Korea Times |
| Total 2027 budget | Nearly 821 trillion won (about $599B) | The Korea Times |
| Status | Cabinet proposal, still needs National Assembly approval | The Korea Times |
What the Cabinet proposal says
On September 1, 2026, the Ministry of Planning and Budget outlined the 2027 budget, a record of nearly 821 trillion won. It designated semiconductors, physical AI and AI data centers as three megaprojects and put 21.3 trillion won behind them, a 97.2 percent increase over this year. That near-doubling matches the "nearly twice the 2026 amount" language in Monday's Reuters headline, and 21.3 trillion won converts to roughly the $15.97 billion figure at a rate around 1,330 won to the dollar. That is our arithmetic, not a statement from either outlet, so treat the match as likely rather than confirmed.
Budget Minister Park Hong-geun laid out the pieces:
- A special account for semiconductors, to expand production capacity, secure core technologies and strengthen the industry ecosystem.
- 2.6 trillion won for physical AI demonstration projects.
- Core infrastructure for all three megaprojects: power, water and industrial complexes.
- 10,000 GPUs, plus data and workers, to help a national AI foundation model reach globally competitive performance and to develop public AI services.
The megaprojects are the first of five investment priorities. The second, at 62.8 trillion won (up 22.7%), covers seven national strategic technologies including fusion energy, small modular reactors, quantum and advanced biotech. The proposal projects total revenue of 880.8 trillion won, up more than 200 trillion won from this year, driven by corporate and income tax from the semiconductor boom.
Where the money comes from: the chip windfall
The funding story is as interesting as the spending. Korea's AI spending is largely paid for by AI demand itself. As Asia News Network reported from Lee's July 13 fiscal strategy meeting, the president proposed a "future response fund" financed by "unprecedented additional tax revenue driven by a semiconductor boom sparked by the AI revolution." He called that revenue "a valuable resource to be deployed during the golden window in which the global race for artificial intelligence supremacy will be decided."
That is a feedback loop worth understanding: hyperscaler and lab spending on high-bandwidth memory lifts Samsung and SK hynix profits, which lifts Korean corporate tax receipts, which the government then recycles into chips, data centers and robots. We traced the supply side of this loop in this week's AI chip supply chain roundup and in our reading of Anthropic's S-1 chip risk factors, which names the memory suppliers every frontier lab leans on.
The flip side is concentration risk. A budget that leans on chip-cycle revenue is exposed if memory prices turn. The government's answer, a dedicated fund to bank excess receipts, is a hedge, but no outlet we reviewed publishes the fund's size or rules yet.
The pledge in context: 2026 versus 2027
Lee made AI a centerpiece of his first budget speech a year ago. In November 2025 he told the National Assembly that 10.1 trillion won (about $7 billion) of the 2026 budget would go toward a push to make Korea one of the world's top three AI powers, and said Korea should lead in physical AI by using its manufacturing capacity and data. The 2027 numbers are a straight continuation: about double, with physical AI and data centers now named as standalone megaprojects rather than line items.
For the wider picture of how Korea got here, read our South Korea AI ecosystem explainer, which covers the chaebol-backed sovereign model stack, HyperCLOVA and the national foundation-model program that the 10,000 GPUs are meant to feed.
Why "physical AI" is the new headline category
Physical AI means systems that let robots, vehicles and industrial equipment perceive and act in the physical world. Korea's case is simple: it has some of the densest factory automation and a leading shipbuilding, auto and electronics base, and it wants to turn that into data for embodied-AI models. The 2.6 trillion won demonstration line is where the government will pay for pilots rather than papers.
Other governments are making similar bets. Germany's pairing with Nvidia is covered in our post on the Germany and Nvidia physical AI partnership reporting, and Korean labs are already shipping open models, such as KRAFTON's Raon-1 mini preview. What is different here is the scale of state money and the explicit link to power, water and industrial-park provision.
The infrastructure bottleneck
Data centers appear in the same sentence as chips and robots, and that matters. Across the US, the biggest fights over AI are now about land, electricity and water, as we documented in the data center backlash in the midterms. Korea's budget answers that in the opposite direction: the state commits to supplying power, water and industrial complexes so projects "proceed according to corporate timetables," in Lee's words from July.
That is a real policy contrast. Where US local governments are adding moratoriums, Seoul is promising to clear the path. The open question is whether Korea's grid and water supply can serve gigawatt-scale AI campuses alongside its existing chip fabs, which are themselves heavy consumers. The June 29 regional plan reported by Asia News Network lists very large corporate investment commitments, including 896 trillion won across the southwestern Honam region led by Samsung Electronics, SK hynix and Amkor, so the demand for utilities will be concentrated and large.
What is confirmed and what is not
- Confirmed by Korean government briefing (as reported by The Korea Times): the three megaprojects, the 21.3 trillion won figure and its 97.2 percent rise, the 2.6 trillion won for physical AI demonstrations, the 10,000 GPUs, and the chip special account.
- Reported by Reuters (headline only for us): the November 2 vow and the $15.97 billion figure, "nearly twice" the 2026 amount.
- Not confirmed: whether the National Assembly will pass the budget unchanged, how the 10,000 GPUs will be procured and from whom, and the size of the future response fund.
- Reported inconsistently: total budget size. The Korea Times cites nearly 821 trillion won ($599 billion), while other July coverage said "over 800 trillion won" and about $530 billion, which reflects different exchange rates and an earlier forecast.
What it means for builders
- Korean-language and sovereign model demand will rise. A named national foundation-model program with guaranteed GPUs means public-sector contracts for domestic labs. If you sell into Korea, expect procurement to favor local stacks.
- Robotics pilots get funded. The physical AI demonstration line is where startups and the chaebol will find paid test deployments.
- Memory remains the choke point. The more governments subsidize compute, the more HBM demand persists. Plan for tight supply and price volatility, and see OpenAI's Samsung custom chip partnership for how labs are hedging.
- Watch the Assembly calendar. Budget bills are amended. If you invest or build on the assumption of a specific number, wait for the final vote.
How to verify
Read the Ministry of Planning and Budget's own release and the Korea Times briefing for the September numbers, then compare with Reuters' full November 2 story for the new remarks. For currency conversions, use the same exchange date for every figure; the gap between $530 billion and $599 billion in coverage of the same budget is largely a rate and forecast difference.
Bottom line
Korea is using a chip-driven tax windfall to buy a national AI stack: chips, data centers, robots and a domestic model. The headline of roughly $15.97 billion is credible against the government's own 21.3 trillion won figure, but it is a proposal, and the real tests are Assembly passage, power and water delivery, and whether the GPUs and demonstrations produce usable systems. We will update this post when the Assembly acts or the full Reuters text and ministry release are available.