South Korea’s government picked a fight with physics on June 29, and the outcome is not guaranteed. President Lee Jae-myung stood alongside Samsung and SK Group executives and unveiled a ₩1,350 trillion investment program, worth roughly $880 billion at the exchange rates cited in later reporting from Reuters and the BBC, meant to keep South Korea’s chip industry competitive through 2035. Al Jazeera went further, describing the combined national AI and chip investment drive as topping $1 trillion once the full ₩2,100 trillion semiconductor-cluster commitment disclosed that same month is factored in.
Three weeks later, Reuters reported the less flattering half of the story. The power grid and water systems that are supposed to feed South Korea’s next generation of chip factories do not exist yet, and building them could take longer than building the fabs themselves. Tom’s Hardware, citing a National Assembly Research Service estimate, put a number on the gap: the country’s flagship Yongin semiconductor cluster alone will need 15 to 16 gigawatts of power at full build, against a local grid that currently supplies about 1.9 gigawatts.
This is the story behind the headline figure. South Korea’s semiconductor investment plan is the largest industrial commitment in the country’s history, and it now runs straight into a power and water problem that no amount of won or dollars can solve on its own.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What South Korea Just Put on the Table
Lee’s June 29 announcement bundled three initiatives into what his government calls the “Three Mega Projects”: semiconductor manufacturing, AI data centers, and robotics or “physical AI.” The package totals ₩1,350 trillion, translated by outlets including the BBC and Bloomberg as approximately About $648 billion, spread across a ten-year window from 2026 to 2035. It is a public-private plan, meaning the figure combines corporate capital from Samsung and SK Group with government infrastructure spending, not a single state budget line.
Lee framed the plan in blunt terms. “Semiconductors, physical AI, and AI data centres are the triple axis for a great leap forward,” he said, according to BBC reporting on the announcement. The pitch is straightforward: South Korea already controls a large share of the world’s memory chip production through Samsung Electronics and SK Hynix, and Lee wants to extend that lead before AI demand consolidates around a smaller set of suppliers.
Breaking Down the $880 Billion
The plan splits into two dominant buckets, with a smaller third component layered on top, one that grew considerably in the weeks after the June 29 unveiling. The Chungcheong chip-packaging cluster started out around ₩81 trillion, per Reuters, but by July 2026 South Korea’s trade ministry told Yonhap that the region’s HBM and packaging buildout had expanded to roughly ₩392 trillion, including a ₩140 trillion Samsung pledge covering a Chungcheong HBM fab plus OLED and display lines, and a ₩100 trillion SK Hynix plan for Cheongju split between an ₩80 trillion M17 NAND fab and ₩20 trillion for P&T7 packaging capacity.
The Fab Half
Roughly ₩800 trillion, or $518 billion to $520 billion depending on the conversion used, goes toward four new semiconductor fabs in South Korea’s southwest, with Samsung and SK Hynix each committing ₩400 trillion toward the buildout, according to Reuters’ reporting on the June 2026 announcement. Samsung and SK Hynix will each build two plants in the Gwangju and Jeolla provinces, an area Al Jazeera describes as previously “neglected” by South Korea’s chip industry, which has historically clustered around Seoul. Industry Minister Kim Jung-kwan confirmed the two companies would build the sites with equipment and materials suppliers attached, according to Al Jazeera’s coverage of the announcement.
The AI Data Center Half
A second ₩550 trillion, close to $356 billion, funds AI data center buildout led by SK Group, GS Group, and Naver. The stated target is 8.4 gigawatts of AI data center capacity by 2029, expanding to 18.4 gigawatts by 2035. That figure alone exceeds the total power capacity of several mid-sized national grids, built for AI training and inference rather than chip fabrication.
| Investment Bucket | Won Value | Dollar Value | Lead Backers | What It Buys |
|---|---|---|---|---|
| Semiconductor fabs (Honam) | ₩800 trillion | ~$518–520 billion | Samsung, SK Hynix | 4 new fabs, 2 each, in Gwangju and the Jeolla provinces |
| AI data centers | ₩550 trillion | ~$356 billion | SK Group, GS Group, Naver | 8.4 GW of capacity by 2029, growing to 18.4 GW by 2035 |
| Chip packaging cluster | ~₩81 trillion | ~$52.5 billion | Chungcheong-region firms | Advanced packaging capacity, per AI-industry tracker aidb.digital |
| Total headline commitment | ₩1,350 trillion | ~$880 billion | Public-private, 10-year horizon | “Three Mega Projects”: chips, AI data centers, robotics |
Samsung and SK also disclosed separate, longer-horizon figures that sit outside this headline number. Samsung told domestic reporters it plans roughly ₩2,450 trillion in total domestic investment between 2026 and 2040, with about ₩2,100 trillion of that earmarked for semiconductor clusters specifically, a figure Reuters says spans at least three separate cluster projects. SK announced a comparable long-term figure of around ₩2,100 trillion for semiconductors and AI data centers combined, and by August 2026 that commitment had started converting into approved spending: SK Hynix’s board signed off on ₩54 trillion for two memory fabs, ₩35.2 trillion for a new Yongin Y2 line and ₩19.1 trillion for the Cheongju M17 facility, according to the company’s own disclosure. Those numbers describe a longer runway than the ten-year “Three Mega Projects” window and should not be added directly on top of the $880 billion figure.
Yongin’s Math Problem: 16 Gigawatts Needed, 1.9 on Hand
The Yongin Semiconductor Cluster, in Gyeonggi Province just south of Seoul, predates the June 29 announcement. Samsung and SK Hynix have been building there since 2023, and it is the project most likely to determine whether South Korea’s broader chip strategy holds together. According to Tom’s Hardware’s review of a National Assembly Research Service report, Yongin will require 15 to 16 gigawatts of electricity at full operation, close to 25% of Seoul metro power demand. Local supply currently sits around 1.9 gigawatts.
Korean financial outlets frame the shortfall a little differently but arrive at a similar warning. The Korea Economic Daily reported that, per the Ministry of Climate, Energy and Environment, Yongin needs 14.8 gigawatts and 1.072 million cubic meters of water per day by 2050, an amount of power comparable to ten large nuclear reactors. Split across the site’s two zones, the National Industrial Complex needs 9 gigawatts and the General Industrial Complex needs roughly 6 gigawatts, for a combined 15 gigawatts. As of December 2025, only about 9 gigawatts, or 60%, had been secured, leaving a 40% gap unresolved.
Samsung is not waiting on the grid alone. The company is building six liquefied natural gas power plants to cover 3 gigawatts of its 9-gigawatt priority allocation, and on July 22 the government said it would compress the timeline for securing Yongin’s remaining power, aiming to deliver most of the needed capacity through LNG well before the site’s original 2039 to 2050 horizon, according to the Korea Economic Daily’s account of the policy shift. Regulators have also closed the door on an alternative fix. When some lawmakers floated relocating part of the cluster to ease the strain, an Industry Ministry official told the Korea JoongAng Daily that “the proposal is not under consideration,” while conceding that “detailed power-supply plans beyond 2039 have not yet been finalized.”
Honam Enters the Race for Power and Water
The new Honam semiconductor cluster, the southwestern hub tied directly to the June 29 announcement, carries its own, separate infrastructure bill. South Korea’s Minister of Climate, Energy and Environment pledged the government would secure 6.3 gigawatts of electricity and 650,000 tons of water per day for factories in the region, according to the Chosun newspaper’s coverage of the announcement. Reuters, which broke the follow-up story on the plan’s infrastructure risk on July 21, reported separately that the government wants the Honam hub operational within four years and that local opposition to its power and water demands ranks among the biggest obstacles to that timeline.
Put the two clusters side by side and the scale of the ask becomes clearer.
| Cluster | Location | Power Needed at Full Build | Power Secured or Available | Water Needed | Target Timeline |
|---|---|---|---|---|---|
| Yongin Semiconductor Cluster | Gyeonggi Province (near Seoul) | 15–16 GW (~25% of Seoul metro demand) | ~1.9 GW local grid, ~9 GW (60%) committed as of Dec. 2025 | 1.072 million m³/day by 2050 | Full build-out by 2050, power fast-tracked toward 2030 |
| Honam Semiconductor Cluster | Gwangju, North & South Jeolla | 6.3 GW | Government pledge of “timely, uninterrupted” supply, not yet built | 650,000 tons (~m³) per day | Hub targeted within 4 years, roughly 2030 |
Combined, the two clusters are asking for more than 21 gigawatts of new or upgraded electricity capacity, on top of whatever South Korea’s existing grid already carries for homes, offices and every other industry. That is before counting the 8.4 to 18.4 gigawatts targeted for AI data centers under the same plan.
Why Water Matters as Much as Electricity Here
Chip fabrication is a water-intensive process. Wafers need ultrapure water for cleaning and etching at nearly every stage of production, and a single large fab can consume millions of gallons a day. South Korea’s plan asks two regions to supply water at a scale the country has not previously attempted for industrial use, more than 1.7 million tons or cubic meters a day across Yongin and Honam combined once both figures are added together.
Gyeonggi Province, home to Yongin, is already one of South Korea’s more water-stressed regions during dry seasons. The southwestern Jeolla provinces have more available water in principle, which is part of why the government chose the region for new fabs, but Reuters reported that local resistance to large-scale water allocation is a live political issue, not a settled one. Energy analysts have raised similar doubts about renewables filling the gap. One government advisor, cited by Chosun Biz, argued that covering Yongin’s 15 gigawatt requirement with solar power alone would require roughly 97 gigawatts of installed solar capacity, given South Korea’s average solar utilization rate of about 15.4%.
The Nvidia Connection Nobody’s Pricing In Yet
South Korea’s chip plan is framed domestically as a manufacturing and jobs story, but its real relevance to the rest of the tech industry runs through a narrower product category: high bandwidth memory, or HBM. Samsung Electronics and SK Hynix are two of only three companies in the world, alongside Micron, that manufacture HBM at the scale modern AI accelerators require. Nvidia’s current Blackwell-generation chips and its upcoming Rubin architecture both depend on HBM3E and next-generation HBM4 stacks sourced primarily from these Korean suppliers.
That dependency is precisely what makes the Yongin and Honam power questions relevant outside South Korea. If new fab capacity comes online without the electricity to run it at full utilization, HBM output growth slows, and Nvidia, AMD and every other AI accelerator maker competing for allocation feels it downstream in the form of tighter memory supply and firmer pricing. This site has covered how AI-driven memory demand is already reshaping prices across consumer gaming hardware, and South Korea’s fab buildout is the supply-side response to that same demand curve. It is also why chip supply chain security has become a recurring flashpoint well beyond South Korea’s borders.
Seoul’s Bet Next to Washington’s and Taipei’s
Measured against other governments’ semiconductor pushes, South Korea’s $880 billion figure is large by any standard. The United States’ CHIPS and Science Act, signed into law in August 2022, appropriated $52.7 billion specifically for semiconductor manufacturing, research and workforce development, part of a broader $280 billion package that included tax credits and science funding. Seoul’s new commitment is roughly 16.7 times that semiconductor-specific figure.
TSMC’s Arizona buildout offers a closer comparison, since it is a single company’s fab investment rather than a national subsidy program. TSMC’s total announced Arizona commitment reached $165 billion after a March 2025 expansion added $100 billion to its original $65 billion pledge. South Korea’s $880 billion is about 5.3 times that figure, though the comparison is not apples to apples. TSMC’s number is corporate capital expenditure for a handful of fabs in one US state, while South Korea’s figure blends multiple companies, a national government, and infrastructure spending across two regions and an entire data center buildout.
What the comparison does establish is scale of ambition. No other government has tied its industrial strategy this explicitly to owning both ends of the AI hardware stack, memory fabrication on one side and data center capacity on the other, inside a single package.
From $450 Billion to $880 Billion in Five Years
This is not South Korea’s first large semiconductor pledge, and the trend line is worth noting. The government previously announced a roughly $450 billion chip-industry blueprint in 2021 and a $400 billion follow-up in 2023, both aimed at strengthening the country’s position against Taiwan’s TSMC and China’s fast-growing domestic chipmakers. As recently as January 2026, the Ministry of Science and ICT was still operating at a far smaller scale, allocating ₩235.1 billion, including ₩24 billion in newly added funding, to core semiconductor, display and battery source-technology projects. The 2026 plan does not simply extend those efforts. At $880 billion, it exceeds the combined total of both prior packages many times over.
The acceleration reflects how quickly the underlying market shifted. The 2021 and 2023 plans were written before generative AI created sustained, structural demand for HBM at current volumes. By 2026, memory chips are no longer just a cyclical commodity business for Samsung and SK Hynix. They are strategic infrastructure, which is part of why this plan places semiconductor fabs and AI data centers inside the same ten-year package rather than treating them as separate policy tracks.
Politics on the Factory Floor
Site selection is rarely just an engineering decision, and this plan is no exception. Choosing Gwangju and the Jeolla provinces for new fabs spreads South Korea’s chip industry beyond its traditional Seoul-Gyeonggi base for the first time at this scale, a shift regional officials have wanted for years. Reuters reported that the government intends to establish the Honam hub within four years specifically to prove the southwest can support advanced manufacturing, not just legacy industry.
That regional expansion adds execution risk on top of the power and water questions. Building transmission lines, water treatment capacity and skilled-labor housing in a region that has never hosted fabs at this scale takes time that a four-year target may not fully accommodate, particularly if local communities push back on the same power-line and water-allocation fights already playing out around Yongin.
What South Korea’s Leadership Is Saying
Lee has been consistent in linking the plan’s urgency to South Korea’s existing infrastructure limits rather than presenting it purely as an expansion story. “Existing facilities based in Yongin and Pyeongtaek have reached their operational limits,” he said, according to TechCrunch’s report on the announcement, explicitly acknowledging that South Korea’s most advanced existing chip hub cannot absorb much more growth without new capacity elsewhere.
He has also cast the plan in geopolitical terms rather than purely economic ones. Bloomberg and other outlets reported Lee describing the investment as a “national survival strategy” for the AI era, language that matches his framing of semiconductors, physical AI and AI data centers as what he called, per the BBC, “the triple axis for a great leap forward.” The rhetoric signals that Seoul views this less as an industrial policy update and more as a bet the country cannot afford to lose to Taiwan, China or the United States.
Market Impact: Memory Prices and AI Hardware Costs
The near-term market effect of South Korea’s plan is likely to run counter to what the headline investment figure implies. New fab capacity takes years to design, permit, build and ramp to volume production, while capital committed to construction is capital not spent expanding output at existing lines. In the near term, that dynamic tends to keep memory supply tighter rather than looser, reinforcing the DRAM and HBM pricing pressure already visible across the PC and AI hardware markets this year.
The knock-on effects extend well past chipmakers. Data center operators planning multi-year capacity expansions, a trend this site examined in coverage of record data center leasing activity, are effectively competing with South Korea’s own AI data center buildout for the same limited pool of advanced memory and power-hungry infrastructure. GPU makers reliant on Nvidia’s Blackwell and Rubin platforms face a similar squeeze, since HBM allocation constraints ripple directly into finished accelerator availability, a dynamic already visible in board-level products like Nvidia’s RTX Spark launch and its DGX Spark platform.
Samsung and SK Hynix vs. Micron and TSMC
South Korea’s plan is, in part, a defensive move against competitors who are not standing still. Samsung and SK Hynix currently anchor global HBM production, but neither company has the market to itself, and the new investment is designed to widen the gap rather than simply hold current share.
Micron and TSMC’s Countermove
Micron remains the third major HBM producer and has been expanding its own US and Asian capacity, while TSMC’s $165 billion Arizona commitment secures advanced logic manufacturing on American soil even though TSMC does not compete directly in HBM production. The more direct competitive pressure on South Korea comes from China, where domestic memory makers such as CXMT have been scaling conventional DRAM output, and from the broader question of whether Nvidia and other AI chip designers diversify their HBM sourcing enough to reduce reliance on any single country. Caixin Global’s reporting on the plan frames it explicitly as a response to that competitive pressure from China’s expanding chip ambitions.
Can Seoul Actually Deliver On Time
The honest answer, based on Yongin’s track record, is not yet clear. That cluster has been under construction since 2023, and as recently as December 2025 only 60% of its required power was secured, according to Korean financial press. The government’s July 22 decision to fast-track the remaining supply by roughly a decade is a meaningful policy shift, but it is also an admission that the original timeline was not working.
Honam starts from an even earlier stage. Ground has not been broken on the scale the plan envisions, the four-year target is aggressive by South Korean infrastructure standards, and Reuters’ reporting suggests local opposition over power lines and water rights is already forming. South Korea has successfully built world-class chip clusters before, but never two at this combined scale on this compressed a timeline.
Five Predictions Through 2030
- More emergency power announcements are coming. Expect additional fast-track policy moves similar to the July 22 Yongin decision, likely including new LNG plant approvals and possibly nuclear restart or extension debates.
- Memory prices stay firm into 2027. New fab capacity will not meaningfully add to global HBM and DRAM supply until construction, equipment installation and yield ramp are complete, a multi-year process that keeps near-term supply tight.
- Nvidia and other AI chip designers deepen direct ties with Samsung and SK Hynix. Long-term supply agreements or co-investment structures become more attractive as a way to lock in HBM allocation ahead of competitors.
- The Honam four-year target slips. Given Yongin’s history of delayed power buildout, a similar delay at the newer, less-developed Honam site is more likely than an on-schedule delivery.
- South Korea leans on this plan diplomatically. Expect Seoul to reference the $880 billion commitment in trade and tariff negotiations with Washington as evidence of its strategic alignment on chip security.
Frequently Asked Questions
What is South Korea’s $880 billion chip and AI plan?
It is a ten-year, public-private investment program announced June 29, 2026, by President Lee Jae-myung, covering semiconductor fabs, AI data centers and robotics under the banner “Three Mega Projects.” The total is reported as ₩1,350 trillion, or approximately $880 billion.
How much power does the Yongin semiconductor cluster need?
Estimates range from 14.8 to 16 gigawatts at full build-out, roughly a quarter of Seoul metropolitan area power demand, against a local grid supply of about 1.9 gigawatts and roughly 9 gigawatts formally committed as of late 2025.
What is the Honam semiconductor cluster?
Honam is the new southwestern chip hub in Gwangju and the Jeolla provinces, where Samsung and SK Hynix will each build two fabs as part of the June 2026 plan. It needs 6.3 gigawatts of power and 650,000 tons of water per day, with a four-year construction target.
How does South Korea’s plan compare to the US CHIPS Act?
South Korea’s $880 billion figure is roughly 16.7 times the $52.7 billion the US CHIPS and Science Act appropriated specifically for semiconductor manufacturing, research and workforce programs in 2022.
Why does this matter for Nvidia and AI chips?
Samsung and SK Hynix are two of the world’s three major producers of high bandwidth memory, which Nvidia’s AI accelerators depend on. Delays to South Korean fab power supply can tighten global HBM availability and pricing.
When will the new chip fabs be finished?
Honam’s hub is targeted for roughly four years from the June 2026 announcement, putting completion near 2030. Yongin’s full build-out, including power infrastructure, has historically been projected out toward 2050, though the government is now trying to compress that timeline.
Who is paying for South Korea’s semiconductor investment?
The plan blends private capital from Samsung, SK Group, GS Group and Naver with South Korean government infrastructure spending, structured as a public-private partnership rather than a single state budget allocation.
Could power shortages delay South Korea’s chip plan?
Yes. Independent reporting from Reuters and Tom’s Hardware both identify power and water infrastructure, not capital availability, as the plan’s biggest execution risk, particularly at the Yongin cluster where only 60% of required power was secured as of December 2025.
