Samsung Electronics plans to spend more than 110 trillion won, or roughly $73.24 billion, on semiconductor capital expenditures and research during 2026. The announcement, made from Seoul on March 19 and first reported by Bloomberg, marks the largest single-year semiconductor investment by any company in history. It exceeds TSMC's estimated $45 billion capex for the same year and dwarfs Intel's reduced foundry budget.
The move comes at a moment when Samsung faces pressure on two fronts. Its foundry market share fell to 7.2% in 2025 while TSMC held 69.9%, and in high-bandwidth memory it has trailed SK Hynix in securing Nvidia qualification. The $73 billion is Samsung's attempt to reverse both trends at once, across memory, foundry, and advanced packaging.
The stakes are plain. The global semiconductor industry is projected to exceed $700 billion in annual revenue by 2027, and the AI accelerator market is driving demand for both logic and memory at a pace the supply chain hasn't matched. If Samsung cannot convert this spending into qualified capacity and competitive yields, the gap with TSMC and SK Hynix hardens into something permanent.
The Numbers Behind the Bet
Samsung's 110 trillion won figure covers capital expenditures across fabrication facilities, memory production lines, advanced packaging infrastructure, and R&D on next-generation process nodes. It represents a 128% increase over the company's 2025 semiconductor capex of 47.5 trillion won. TrendForce found that Samsung and SK Hynix combined spent 43.198 trillion won on chip facilities in the first half of 2026 alone, up 35.1% year-over-year.
"We plan to execute investments exceeding 110 trillion won this year to secure leadership in the AI semiconductor era," Samsung said in its March 19 announcement. The company named AI accelerator memory, advanced foundry nodes, and next-generation packaging as the three pillars of the plan.
The spending is funded by a financial turnaround. Samsung's Device Solutions division posted Q4 2025 revenue of 44.0 trillion KRW, up 46.2% year-over-year, with operating profit of 16.4 trillion KRW, a 465% increase over Q4 2024. Full-year 2025 semiconductor revenue reached 130.1 trillion KRW with operating profit of 24.9 trillion KRW. On October 8, 2026, Samsung reported a near-ninefold year-over-year quarterly operating-profit increase driven mainly by AI-memory demand.
Dan Hutcheson, Vice Chairman of TechInsights, put the scale in perspective. "The cost of staying competitive in leading-edge chips has reached a point where only three or four companies on the planet can even attempt it," he said. "Samsung is signaling that it intends to remain one of them, regardless of the short-term financial pain."
HBM4: Samsung's Path Back Into Nvidia's Supply Chain
Samsung has lagged SK Hynix in securing Nvidia's HBM qualification, a gatekeeper role that determines which memory suppliers can serve the most lucrative AI chip market. The company's 12-layer HBM3E stack struggled through much of 2025 and secured approval only for select Nvidia accelerators sold in China.
HBM4 is the reset. Samsung sent samples to Nvidia in September 2025 and entered final qualification. Mass production began in February 2026, and the company has sold out its entire 2026 HBM4 capacity, according to Korea Economic Daily. Samsung also displayed HBM4E samples at SEDEX 2026 on October 14, so HBM4 is no longer the newest generation.
Mark Lipacis, semiconductor analyst at Jefferies, called the qualification a genuine reset. "If their HBM4 performs as early reports suggest, Samsung could recapture significant share from SK Hynix in the next-generation memory cycle," he said. The $73 billion investment ensures Samsung has the manufacturing capacity to deliver on those orders.
The HBM market is expected to exceed $30 billion in 2026. Samsung's sold-out capacity suggests it will capture a meaningful share of that, though the HBM3E experience is a reminder that qualification is not a single event. It is a moving target that Nvidia tightens with each generation.
The Foundry Problem: A 62-Point Gap
Samsung's foundry business tells a harder story. TSMC captured 69.9% of the global foundry market in 2025 with $122.54 billion in revenue. Samsung was second at 7.2% with $12.63 billion, a figure that declined 3.9% year-over-year. The gap between the two widened to 62.7 percentage points.
In Q3 2025, TSMC held 71% while Samsung slipped to 6.8%. SMIC reached 5.3% by Q4 2025, threatening Samsung's second-place position. These figures represent a steep fall from Samsung's 13% foundry share in Q1 2024.
The technology gap is the core issue. TSMC's yields on 3nm and emerging 2nm nodes have consistently outperformed Samsung's gate-all-around transistor technology. Apple, Nvidia, AMD, Qualcomm, and Broadcom overwhelmingly choose TSMC for their most advanced products. Samsung's $73 billion includes substantial allocation for its 2nm GAA process, designated SF2 and SF2P, and the company has secured some 2nm business from Tesla. But competitive yields remain unproven at scale.
Taylor, Texas: The American Fab Gambit
Samsung's $37 billion Taylor, Texas fab is central to its foundry recovery plan. Operations are scheduled to begin in 2027, although available reporting does not confirm an exact opening date. Samsung had earlier targeted operational readiness by the end of 2026, with risk production in the second half of that year, and industry reports from March 2026 suggested full-scale mass production could slip to early 2027.
The fab is designed for Samsung's most advanced process nodes and is positioned to serve U.S. customers seeking domestic manufacturing. It also gives Samsung a stake in the reshoring push that has reshaped semiconductor policy in Washington.
The timing risk is real. A slip into 2027 means Samsung enters the next foundry cycle behind TSMC's Arizona ramp and Intel's 18A push. The money is committed either way.
A Longer Horizon: The $1.8 Trillion Framework
The $73 billion figure is only the first year. On June 30, 2026, Samsung and five affiliates announced a combined domestic investment framework of 2,450 trillion won, roughly $1.8 trillion, running from January 2026 through December 2040. Reporting through October 2026 describes multiple separate projects rather than one independently confirmed $73 billion program, so the headline figure is best read as a reported 2026 company-wide estimate rather than a single commitment.
Samsung also broadened beyond semiconductors. On September 30, 2026, Samsung Electronics and five affiliates invested a combined $1 billion in AI-infrastructure company Helix, including $500 million from Samsung Electronics. The bulk of spending remains aimed at chips, where competition is fiercest.
What It Means for the Industry
Samsung's investment reframes the semiconductor landscape in one respect: the cost of entry at the leading edge has risen to a level only a handful of companies can afford. Whether Samsung converts that spending into market share is a separate question.
TSMC's dominance rests on a virtuous cycle of volume, yield improvement, and customer wins that Samsung has not replicated. Memory is Samsung's stronger hand, and HBM4 qualification is the clearest near-term catalyst. The foundry recovery depends on SF2 and SF2P yields and on the Taylor fab opening on schedule.
It is not clear this will hold. Samsung has made big bets before and lost ground. The difference in 2026 is that the AI memory market is large enough to fund the attempt, and the company's Q4 2025 earnings gave it the cash to try. Whether that cash turns into competitive silicon is the question that matters, and the answer will not arrive before 2027.
For more on foundry market dynamics, see Reuters coverage of semiconductor manufacturing. For related coverage on this site, see desk coverage.