Samsung

Samsung operates the world's second-largest semiconductor foundry and manufactures memory, competing with TSMC and SK Hynix. Struggles with advanced-node yields vs TSMC; strong in HBM and DRAM. Gate-all-around (GAA) transistors shipping at 3nm.

Samsung is the only other foundry pursuing leading-edge logic at scale, critical for supply-chain resilience in AI chips.

samsung.com Foundries

South Koreapublic · 005930.KS

At a glance

Total raised

Undisclosed

no public funding on record

Headcount

—

no public estimate on record

Headquarters

Suwon

South Korea

Founded

1983

43 years active

Status

Public

Ticker · 005930.KS

Patents on file

USPTO grants

For subscribers

Subscribe to see the full profile

Including Funding rounds, Bull / Bear thesis, Stock + earnings, Roster changes, Patents, News, and Open roles.

Founding

50% off

/month

91 of 100 spots left

Full

$10

/month

Available once all 100 Founding Member spots are claimed.

Get full access

Already subscribed? Sign in →

Funding history

Total raised

—

Rounds tracked

Latest stage

IPO

Date	Stage	Amount	Valuation	Lead investors
Jun 11, 1975	IPO	—	—	—

Stock performance

How they make money

Hardware (one-time)Pattern

Samsung runs a hardware manufacturing model, selling physical semiconductor products — memory chips (DRAM, NAND, HBM) and foundry-fabricated logic chips — to OEMs, cloud hyperscalers, and electronics manufacturers. Revenue is recognized when finished wafers are tested, diced, and shipped to customers; most sales are structured as purchase orders for discrete batches rather than recurring contracts. Within the foundry segment, customers tape out designs and pay per wafer lot, with revenue tied to volume output and yield rates. The memory side is more commoditized, driven by spot and contract pricing in a cyclical market. What scales is fabrication capacity: each new fab adds fixed cost but unlocks incremental wafer starts. Gross margin in foundry depends heavily on yield at advanced nodes (sub-7nm), while memory margins swing with supply-demand balance. Both segments are capital-intensive — Samsung spends tens of billions per cycle on fabs and equipment — so depreciation is the dominant cost line. There is no software-like recurring revenue or take-rate; the business is purely about selling higher-value silicon per wafer over time.

Generated by deepseek-v4-flash · June 2, 2026Analytical summary · not investment advice

AI bull / bear

Bull case

Samsung is the only credible alternative to TSMC for leading-edge logic, and geopolitical pressure for supply-chain resilience is a tailwind that plays directly into its hand. Gate-all-around (GAA) transistors already shipping at 3nm give Samsung a structural technology differentiator — TSMC is still on FinFET at that node. If GAA yields mature, Samsung leapfrogs in power and performance for AI accelerators. Meanwhile, Samsung's dominance in HBM3E memory (stacked DRAM for Nvidia and AMD GPUs) means it captures AI spend from two directions: logic wafers and high-bandwidth memory. The vertical integration play — logic, memory, and advanced packaging under one roof — is something TSMC cannot replicate. Recent Samsung research contributions on memory reliability signal deepening technical moats in the AI stack (source_article_ids:). As hyperscalers demand second sources, Samsung is the only foundry scalable enough to answer. Samsung's foundry business is caught between a dominant incumbent and a capital-intensity trap. TSMC holds roughly 90% of leading-edge market share and consistently delivers better yields at every node — customers simply cannot afford the risk of Samsung's reported sub-50% 3nm GAA yields. The GAA "first-mover" advantage has backfired: Samsung rushed to market while TSMC waited to perfect the transistor, meaning early adopters may flee back to TSMC's 3nm FinFET (which performs competitively) or wait for TSMC's own GAA at 2nm. The foundry division is also chronically under-invested relative to Samsung's memory cash cow — capital allocation favors DRAM/NAND, starving logic of the R&D needed to catch TSMC. Without a marquee external customer like Nvidia or Apple committed to Samsung's advanced nodes, the foundry risks becoming a captive supplier for Samsung's own system LSI division, never achieving the scale needed to compete.

Bear case

Samsung's foundry business is caught between a dominant incumbent and a capital-intensity trap. TSMC holds roughly 90% of leading-edge market share and consistently delivers better yields at every node — customers simply cannot afford the risk of Samsung's reported sub-50% 3nm GAA yields. The GAA "first-mover" advantage has backfired: Samsung rushed to market while TSMC waited to perfect the transistor, meaning early adopters may flee back to TSMC's 3nm FinFET (which performs competitively) or wait for TSMC's own GAA at 2nm. The foundry division is also chronically under-invested relative to Samsung's memory cash cow — capital allocation favors DRAM/NAND, starving logic of the R&D needed to catch TSMC. Without a marquee external customer like Nvidia or Apple committed to Samsung's advanced nodes, the foundry risks becoming a captive supplier for Samsung's own system LSI division, never achieving the scale needed to compete.

Generated by deepseek-v4-flash · June 2, 2026Refreshes quarterly · see disclosures

In the news

Semiconductor Engineering6/2/2026
Chip Industry Technical Paper Roundup: Jun. 2
Micron, Samsung, and AMD contribute to semiconductor research papers on inference scaling, memory reliability, and GPU power management.

Watch

Leadership

Jun Young-hyun

Vice Chairman & CEO, Head of Device Solutions (DS / Semiconductor) Division

—

Roh Tae-moon

CEO, Head of Device eXperience (DX) Division

—

Products

SF2 (2nm) GAA Foundry Process
Samsung Foundry's 2nm-class node built on Gate-All-Around (MBCFET nanosheet) transistors, which Samsung adopted ahead of TSMC at 3nm and now extends to SF2. It improves performance, power, and area over SF3 and adds an Extreme High Density (xHD) library to shrink area further, while staying design-compatible with SF3. SF2 is central to Samsung's bid to close the leading-edge foundry gap with TSMC and win AI and HPC silicon.
HBM3E 'Shinebolt'
Samsung's high-bandwidth memory for AI accelerators, branded Shinebolt. The 12-layer (36GB) stack delivers up to ~1,180 GB/s of bandwidth at 9.2 Gbps per pin. Shinebolt is Samsung's entry in the fierce HBM race against SK Hynix and Micron to supply memory for Nvidia and other AI chips, an area where Samsung has worked to requalify its parts and regain share.
HBM4
Samsung's next-generation HBM, which entered mass production in early 2026 after qualification sampling to Nvidia. Built on advanced 1c DRAM with a 4nm logic base die, it delivers up to ~2.7x higher throughput and ~40% better power efficiency than HBM3E, with a 2,048-bit interface enabling up to ~3,300 GB/s. Samsung reportedly sold out its entire 2026 HBM4 capacity, underscoring the memory crunch driving the AI buildout.
DDR5 DRAM & V-NAND Flash
As the world's largest memory maker, Samsung supplies leading-edge DDR5 and LPDDR DRAM on its 1b/1c nodes plus high-layer-count V-NAND flash for SSDs across data-center, mobile, and PC markets. This commodity-and-premium memory franchise, alongside HBM, makes Samsung a primary supplier of the DRAM and storage that AI servers and consumer devices depend on.

Most recent patents

31 patents on file, but none with both an extractable figure and an abstract on Google Patents yet.