Signage of Samsung Electronics is displayed outside the company’s Seocho building in Seoul on July 30, 2026.
Jung Yeon-je/AFP via Getty Images
South Korean President Lee Jae-myung met with Samsung Electronics Chairman Lee Jae-yong on August 26 to discuss chips — the latest in a rapid-fire series of presidential summits that has become Seoul’s primary tool for keeping the country’s ₩800 trillion (approximately $578 billion) Honam chip plan on track amid mounting US demands that Korean chipmakers replicate their most advanced manufacturing operations on American soil. The meetings reveal a governance mechanism that no trade announcement has made explicit: President Lee is running a systematic chaebol summit circuit — Samsung this week, SK Group Chairman Chey Tae-won two weeks before, Hyundai Motor Group Executive Chair Euisun Chung expected imminently — to personally review each conglomerate’s investment commitments and prevent capital from migrating toward the United States under Washington’s pressure.
What that circuit must navigate is a structural paradox the article’s framing rarely names: Washington is simultaneously the demand-side pressure (insisting Samsung and SK Hynix build front-end memory wafer facilities on American soil) and the supply-side gatekeeper (the Gwangju Air Base, a US-Korea joint military facility, sits on the only flat, infrastructure-ready parcel identified for the Honam cluster, and cannot be cleared for fab construction without US bilateral agreement). Seoul cannot build its chip plan at home without Washington’s consent, and Washington is making that consent contingent on Korean chipmakers reducing their production footprint at home. That is the bind President Lee’s summit circuit exists to manage.
Presidential Summits as a Coordination Mechanism, Not a Photo Opportunity
The Lee-Lee Jae-yong meeting on August 26 was explicitly framed by Korean government and business sources as a follow-on to the June 29 Three Mega Projects briefing, where Lee sat alongside both Samsung and SK Hynix chairmen to announce the government’s sweeping public-private framework. The Kyunghyang Sinmun reported that the series of meetings is seen as a venue to review the investment plans of Korean companies amid the race for AI supremacy and to hear voices from the field — not merely to take photographs.
The distinction matters. Lee’s approval rating has fallen to 40.2% recently, the sixth consecutive week of decline, driven by domestic controversies over housing, prosecution reform, and military matters. The conglomerate summit circuit is in part a political maneuver: Lee’s core industrial megaprojects — the Honam semiconductor cluster, AI data centers, physical AI manufacturing — are among the few agenda items with genuine cross-party and cross-class support. Demonstrating active, in-person engagement with Samsung’s chairman reinforces the administration’s message that the chip plan is not a political abstraction but a managed, implementation-stage commitment.
Beyond politics, business sources cited by multiple Korean outlets emphasize that these meetings are genuinely operational. Lee has urged companies to invest domestically and has signaled the government is exploring regulatory relief and other policy steps to create a more favorable business environment. The outcome of Wednesday’s Lee-Lee Jae-yong session was not immediately disclosed, but the agenda — as reported by Korean government and business sources — centered on Samsung’s semiconductor and industrial issues, and specifically on the competing demands of Honam expansion and US fab pressure.
A ₩800 Trillion Belt Across the Southwest
At the center of Wednesday’s discussions was the Honam semiconductor cluster — an ₩800 trillion (approximately $578 billion) project to build four new advanced fabrication plants in the southwestern Gwangju and South Jeolla Province area. Samsung Electronics and SK Hynix committed to four Honam fabs on the designated site, which covers approximately 8.26 million square meters surrounding Gwangju Air Base.
The scale is difficult to overstate. Samsung’s broader participation in the Three Mega Projects involves a decade-long group-wide commitment of approximately ₩1,000 trillion ($722 billion) across semiconductors, AI data centers, and physical AI manufacturing. Samsung’s previously announced five-year domestic investment program — ₩450 trillion (approximately $325 billion) — runs through 2030 and includes a new production line at Pyeongtaek expected to begin operations in 2028. Industry analysts note that it remains unclear whether all reported figures represent new capital or include previously announced commitments.
Officials set an ambitious construction timeline: groundbreaking in late 2026 or early 2027, initial power and water infrastructure operational by 2029, and mass production underway by 2030. That compressed schedule is considerably more aggressive than historical norms for projects of this magnitude — a compression the government has sought by promising administrative fast-tracking, rapid permitting, and infrastructure support.
The rationale for the southwestern location is partly industrial and partly political. The Honam region offers high electricity self-sufficiency relative to the Seoul metropolitan area, abundant water resources, and an established research base anchored by Chonnam National University and the Gwangju Institute of Science and Technology. Critically, it also serves Lee’s commitment to balanced regional development in a country where economic activity is historically concentrated north of Seoul — and the Honam region is the traditional electoral stronghold of Lee’s Democratic Party. Opposition lawmakers from the People Power Party have argued that the site selection was driven more by regional electoral strategy than by industrial logic.
Washington Squeezes and Controls: The Gwangju Paradox
The structural entanglement that the summit circuit is trying to manage runs deeper than the financial strain of simultaneous domestic and overseas fab commitments. Gwangju Air Base serves as a South Korea-US Air Force collocated operating base — a facility where US air assets can deploy during wartime, and where reserve supplies and equipment for combat scenarios are maintained under the Status of Forces Agreement even during peacetime. The base is the designated foundation of the Honam cluster.
Korea-US bilateral consultations on relocation began in July 2026. The US Seventh Air Force has confirmed it has “important military interests at Gwangju Air Base” and is coordinating closely with the Republic of Korea Air Force. As of August 10, no relocation agreement had been reached. The 8.26-square-kilometer (3.19-square-mile) site cannot begin semiconductor development until that bilateral agreement is in place.
Compounding this is an unresolved domestic obstacle: Muan County — the preliminary relocation site for the displaced military airport — had opposed the project and boycotted the official site selection committee’s first two meetings. Muan County agreed to attend the site selection committee for the first time on August 28 — following negotiations that narrowed their preconditions (including demands for ₩1 trillion in government and special city support and designation as a national industrial complex) — a potential turning point that would allow formal selection procedures to advance.
This means that Washington is not a passive observer in the Honam cluster’s domestic timeline. By controlling the pace of bilateral base-relocation talks, Washington effectively co-authors when ground can break. And Washington is using its commercial leverage — the threat of memory tariffs — simultaneously. That dual control is the structural paradox Lee’s conglomerate summits cannot resolve through political alignment alone. It requires a bilateral negotiating track that moves in parallel.
What Washington Wants and Why Existing US Commitments Don’t Answer It
US Commerce Secretary Howard Lutnick made Washington’s demand explicit at a Micron Technology facility event in Clay, New York, publicly urging Samsung Electronics and SK Hynix to invest in US memory chip production and telling reporters the two companies would ultimately “have no choice” but to follow Micron’s example. The framing was competitive and pointed: Micron’s $250 billion domestic commitment over 10 years through 2035 was presented as the model Korean rivals should replicate.
The demand cuts to a technical distinction that is rarely explained in diplomatic coverage. A front-end memory wafer fab is where memory chips are actually manufactured — where lithography tools pattern circuits onto silicon, where through-silicon via (TSV) etching creates the vertical electrical channels that allow individual memory dies to be stacked up to 12 layers high in a high-bandwidth memory (HBM) package, and where hundreds of precision steps transform raw silicon wafers into functional DRAM dies. A back-end facility takes those dies — already fabricated in Korea — and assembles them into finished modules.
Samsung’s Taylor foundry campus in Texas cost $37 billion and is a front-end facility — but it manufactures logic chips under contract for other companies, not DRAM or NAND flash. SK Hynix’s advanced packaging plant in West Lafayette, Indiana, scheduled for groundbreaking this week, is a back-end facility: it will assemble HBM dies made in Korea’s Icheon and Yongin facilities into finished stacks. Neither plant meets Washington’s demand for the process nodes where DRAM and NAND flash memory dies are grown from blank silicon wafers. The US has zero front-end HBM wafer fabrication today.
Building such a facility on US soil is not merely expensive — it is structurally difficult in ways the price tag alone cannot capture. A leading-edge DRAM wafer fab requires 200 to 400 megawatts (MW) of continuous power — equivalent to a mid-sized city — as well as tens of millions of gallons per day of ultra-pure water and a specialty chemical supply chain (hydrogen fluoride, nitrogen trifluoride, specialty photoresists) currently produced almost entirely in Japan and South Korea. Construction costs in the US run approximately double those in Korea. And the trained process engineering workforce required is concentrated in Icheon, Gyeonggi Province, and Hwaseong — not in any US metropolitan area. SK Group Chairman Chey Tae-won acknowledged these constraints directly in a CNBC interview in August: “I’m willing to do it, but finding the right place is really difficult.”
Even if Samsung or SK Hynix committed today to a front-end US fab, five or more years would elapse before the first commercial wafer — meaning the output would not arrive before the early 2030s, well after the 2026–2030 AI infrastructure buildout that is the primary demand driver.
HBM4: Why Samsung Needs Every Ton of Capital Right Now
The presidential summit circuit is unfolding at precisely the moment Samsung is trying to win back its position in the AI memory race after an 18-month stumble on its previous-generation HBM3E chips. The company began HBM4 mass production in February 2026 — the first chipmaker in the world to do so — and has since shipped commercial products to customers including Nvidia.
Samsung’s HBM4 implementation uses a 4-nanometer FinFET base die process at the Pyeongtaek P4 facility — the production hub for the AI memory stack that every major accelerator platform requires. The Pyeongtaek campus also hosts the company’s P4 facility, which operates at high utilization prioritizing HBM4 output. The company is targeting an approximately 50% increase in overall HBM production capacity by year-end.
That expansion target is only as good as Samsung’s yield and customer qualification rates — a point analysts have repeatedly stressed. Samsung’s HBM4 yield hit 80% by August 2026, up from roughly 50% during the sample phase earlier in the year, driven partly by advances in the TC-NCF (Thermal Compression Non-Conductive Film) bonding process that previously constrained HBM stacking precision. UBS projected in August that Samsung narrows HBM share gap in 2027, with Samsung capturing approximately 41% of HBM bit shipments against SK Hynix’s 39%.
The competitive dynamics make capital timing critical. SK Hynix retains an allocation advantage on Nvidia’s current Vera Rubin AI accelerator platform — holding an estimated 60 to 70% of HBM4 volume to Samsung’s roughly 25% — reflecting SK Hynix’s earlier entry into qualification cycles and its established supplier relationship with Nvidia. Any diversion of Samsung’s engineering focus or capital toward US memory fab planning, even on a five-year horizon, risks compounding execution risk at a moment when Samsung is working to reassert its market position with the one technology most central to the AI compute economy.
The Trade Deal’s Narrow Path
The geopolitical context for Lee’s balancing act is the bilateral agreement finalized at the APEC summit in Gyeongju in October 2025 and ratified by the Korean National Assembly in March 2026 — passed by 226 votes to eight after President Trump threatened on social media to reimpose 25% tariffs if the bill stalled.
Under the deal, South Korea agreed to invest $350 billion in US industries: $150 billion into US shipbuilding and $200 billion across strategic sectors including semiconductors, pharmaceuticals, critical minerals, energy, AI, and quantum computing. In exchange, reciprocal tariffs and certain Section 232 tariffs on South Korean goods were capped at 15%. The enabling legislation established the Korea-US Strategic Investment Corporation, authorized to fund up to $20 billion per year in US investments until the full $350 billion commitment is met.
The deal’s internal structure creates the specific tension that Lee’s summit circuit is navigating. Korea’s Industry Minister Kim Jung-kwan completed a two-day Washington investment visit in August, meeting with Lutnick and US Trade Representative Jamieson Greer, and told reporters there had been “significant and meaningful progress.” A formal announcement on the first investment projects under the framework is expected in September. Seoul has officially denied that a semiconductor fab project is being negotiated as the first commitment, with officials pointing to the energy sector as the more likely first candidate. That denial reflects the sensitivity of the semiconductor issue: committing chips as the first investment would signal that Washington’s memory fab demand has effectively been accepted as a priority.
The deal set sector-specific limits on future tariffs for semiconductors subject to further negotiations — including provisions related to the 25% global semiconductor tariff imposed in January 2026. That ongoing ambiguity gives Washington continued leverage. The 15% tariff cap is a guarantee that can be tested against future enforcement actions; the 100% memory tariff threat remains a coercive instrument rather than an enacted policy, and it has a significant structural problem: Samsung and SK Hynix together control approximately 70% of global DRAM output, and there is no near-term US alternative.
Balancing Two Clocks
President Lee’s series of conglomerate meetings reflects a recognition that the trade deal’s investment obligations cannot be discharged without active alignment between government and industry — and that the domestic mega-projects must proceed in parallel rather than yielding to US demands.
Business officials who have spoken to Korean media are candid that individual corporate decisions will not be enough. The balance between domestic production bases and overseas expansion — particularly in memory, where Korea’s technological lead over China’s rising CXMT (ChangXin Memory Technologies) is eroding in commodity DRAM segments — will require policy support through trade negotiations and industrial incentives that go beyond what any conglomerate can negotiate alone. CXMT’s global DRAM market share rose from approximately 3% to 8% in a single year, and its commodity chips are priced 15 to 20% below Korean equivalents; HP and Dell have begun quality-testing its products.
What Lee’s circuit is ultimately running is a two-clock problem: the domestic chip plan runs on a construction and permitting clock (Muan County, Gwangju airbase bilateral clearance, enabling legislation, power and water infrastructure), while the US pressure runs on a political clock (the $350 billion commitment’s September announcement window, the tariff threat, Lutnick’s competitive framing). Neither clock waits for the other. The Honam cluster groundbreaking — expected before the end of this year or in early 2027 — will be the first real test of whether the two can be reconciled.
Hyundai Motor Group meeting expected next, with Executive Chair Euisun Chung, completing a systematic conglomerate review that together spans Korea’s three largest industrial contributors to the three mega-projects.
Exchange rates are approximate, based on ₩1,385 per US dollar as of August 27, 2026.
Frequently Asked QuestionsWhy is President Lee personally meeting each chaebol chief rather than delegating these discussions to his industry minister?
The presidential meetings serve a function that ministerial-level talks cannot: they signal to companies that the domestic mega-projects are a presidential priority, not a bureaucratic process that can be managed at arm’s length. When Lee meets with Samsung’s chairman directly, he is putting the weight of the presidency behind the expectation that Samsung will hold its Honam commitments. That signal matters at a moment when US pressure — backed by tariff threats — is providing a competing incentive for capital allocation. Business sources cited by multiple Korean outlets also note that Lee is receiving direct field intelligence about what companies need from the government to proceed: regulatory relief, permitting acceleration, infrastructure coordination. These are the inputs that translate a mega-project announcement into an operational construction timeline.
Why can’t South Korea simply build the Honam cluster and commit to US memory fabs simultaneously?
The core constraint is not political will — it is capital, engineering bandwidth, and timeline physics. Samsung’s Honam commitment alone runs to ₩400 trillion (approximately $289 billion), on top of its existing Yongin expansion (₩360 trillion, approximately $260 billion) and Pyeongtaek buildout. A US front-end memory fab would cost an additional estimated $15 billion to $20 billion or more at Korean construction cost — double that at US cost levels — and require a trained workforce that does not currently exist in any US location. Even if both commitments could be financed simultaneously, the engineering talent and management bandwidth required to simultaneously execute two of the largest fab projects in semiconductor history would be stretched beyond what either company could realistically sustain in the 2026–2030 window. The first HBM wafers from a hypothetical US front-end fab would not arrive before the early 2030s regardless of when the commitment is made.
What is the significance of the Gwangju Air Base in the Honam cluster timeline?
The 8.26-million-square-meter (approximately 2,042 acres) Gwangju Air Base site is the only parcel identified for the Honam cluster that offers the combination of flat terrain, existing leveled ground, and proximity to the Gwangju urban research ecosystem. Moving the base requires relocating US military war reserve materials and transferring SOFA-designated areas — both of which require bilateral US-Korea agreement. South Korea and the United States began formal consultations in July 2026; as of mid-August, no timeline had been finalized. In parallel, Muan County — the domestic relocation destination — had refused to engage with the site selection process until late August 2026, when a resolution of its preconditions began to narrow the gap. The airbase relocation is not a peripheral issue: until it is resolved, no fab foundation can be dug.
How close is Samsung to qualifying HBM4 at full scale with Nvidia and other AI customers?
Samsung cleared HBM4 qualification with both Nvidia and AMD before the commercial ramp began, resolving the 18-month qualification stumble that cost it the HBM3E supply cycle. However, qualification and volume allocation are different things: Nvidia’s Vera Rubin platform currently allocates an estimated 60 to 70% of HBM4 volume to SK Hynix and roughly 25% to Samsung, reflecting SK Hynix’s earlier entry into qualification cycles and its established supply relationship. Samsung’s path to recapturing share runs through yield stabilization and consistent delivery — its HBM4 production yield reached approximately 80% by August 2026, and UBS projects Samsung will narrow the gap to near-parity with SK Hynix in overall HBM bit shipments by 2027. Whether that projection holds depends on maintaining the yield trajectory through the full production ramp at Pyeongtaek P4 — exactly the kind of focus that a simultaneous US fab commitment would threaten.