brain/
← all entities
entitySSNLFstock-market

Samsung Foundry

Notes

Samsung Foundry

One-line summary: South Korean integrated device manufacturer with a credible 2nm foundry ramp underway — SF2P at 70% yield (January 2026), Taylor Texas fab 90% production-ready (April 2026), and Apple's simultaneous evaluation alongside Intel makes Samsung the third leg of the leading-edge foundry competition.

What it is

Samsung's semiconductor division has two arms relevant here: (1) Samsung Foundry — contract manufacturing at leading nodes (SF2P / GAA-based 2nm); and (2) Samsung Memory (HBM, DRAM). For the us-fab-capacity-bottleneck thesis, Samsung Foundry is the wildcard that could dilute Intel's position as the primary TSMC alternative.

Why it matters to stock-market

Samsung matters in two ways. First, as a competitor to Intel Foundry: if Apple or hyperscalers send leading-edge design wins to Samsung instead of (or alongside) Intel, the Intel re-rate thesis is a smaller-magnitude event. Second, Samsung Memory (SK Hynix competitor) is a downstream indicator of HBM demand — see hbm-supply-bottleneck.

Key facts

  • SF2P yield: 70% as of January 2026 (Financial Content / tokenring, Feb 5). SF2P improvements vs SF2: +12% clock speed, +25% power efficiency, -8% die area. SF2P+ announced for 2026 with further gains; 1.4nm pushed to 2028–2029.
  • 4nm yield (stronger near-term story): Samsung 4nm >80% yield (Digitimes May 4); foundry utilization at "highest level ever" (Q1 2026 earnings April 30). HBM4 base dies on 4nm (Nvidia, AMD, OpenAI). Nvidia AI inference chips (Groq-based, 4nm, H2 2026). Samsung path to profit Q2 2026 after estimated KRW 1T operating loss Q1 2026.
  • Taylor, Texas fab: Q2 2026 initial 2nm risk production commenced February 2026. Targeting operations H2 2026; high-volume production is realistically a 2027 story.
  • High-NA EUV 2026: 2 units — equal allocation to Intel, higher than TSMC (0).
  • Apple-Samsung: Exploratory only; Intel's preliminary Apple deal "puts Samsung foundry ambitions under pressure" (Digitimes May 12, 2026). No orders placed; Samsung is a secondary evaluation, not co-equal to Intel's position.
  • Qualcomm: April 23, 2026 Digitimes — "Qualcomm mulls return to Samsung 2nm as TSMC targets its LPU market." Still not closed. WCCFTech: full flagship switch impossible on timing; dual-source most likely (lower/mid-tier Snapdragon to Samsung, flagship to TSMC). CEO Amon confirmed Samsung talks at CES January 2026. From 2026-05-22-autoresearch-samsung-foundry-may-22-2026.
  • AMD: TrendForce (Dec 2025) — "AMD nears Samsung Foundry deal for 2nm CPUs amid TSMC crunch." Not confirmed as of May 22, 2026. If AMD closes, Samsung gains a major US AI/data center CPU anchor customer. From 2026-05-22-autoresearch-samsung-foundry-may-22-2026.
  • Tesla AI6: Samsung won a $17B deal for Tesla AI6 automotive AI chips at Taylor (TrendForce July 2025). This is separate from the Tesla $3B Austin Intel 14A R&D fab — one program each foundry. Samsung handles automotive inference silicon; Intel handles Terafab-related R&D and future xAI/foundry compute.
  • CHIPS Act: US government subsidies received for Taylor, but not at Intel's $8.9B grant + $11B loan scale. No White House champion at Intel's political level.
  • Q1 2026 overall results (April 30, 2026): Samsung Electronics total revenue 133.9T won; operating profit 57.2T won (beat 55.3T consensus; +48x YoY from ~1.2T won Q1 2025). Device Solutions (DS) division: ~53.7T won op profit — driven overwhelmingly by HBM/DRAM, not foundry. Q1 2026 semiconductor profit exceeds Samsung's entire FY2025 semiconductor total. From 2026-05-20-autoresearch-samsung-foundry-q2-2026-profit-2nm-qualcomm-risk.
  • Foundry FY2026 profit goal: 2 trillion KRW internal target. H2 2026: mass production of SF2P for mobile; 4nm memory products; LPU (large processing unit) chips for AI/HPC. Targeting full utilization of advanced-node production lines in Q2 2026. From 2026-05-20-autoresearch-samsung-foundry-q2-2026-profit-2nm-qualcomm-risk.
  • SF2 vs SF2P yield divergence (critical distinction): SF2P (second-gen 2nm): 70% (January 2026) — the node used for Apple evaluations and high-profile deals. SF2 (first-gen 2nm): ~55% (TrendForce, April 14, 2026) — below mass-production threshold; Qualcomm may revert to TSMC rather than commit Samsung 2nm for upcoming Snapdragon chips. The 70% figure commonly cited refers to SF2P only; the mass-market SF2 node is still below threshold. From 2026-05-20-autoresearch-samsung-foundry-q2-2026-profit-2nm-qualcomm-risk.
  • Samsung Memory — Nvidia HBM qualification CLEARED at scale: Samsung passed Nvidia's 12-layer HBM3E qualification (Sept 2025), began Nvidia HBM3E shipments in Q3 2025, and sold out its 2026 HBM (preorders exceeding planned output); its HBM4 reportedly passed Nvidia's tests and it shipped industry-first HBM4E samples. HBM revenue share ~22% (Counterpoint Q3 2025) behind SK Hynix ~57%, ahead of/level with Micron ~21%. This resolves the long-standing "Samsung struggles with Nvidia qualification" tension in Samsung's favor — the HBM supplier market is now genuinely three-way competitive, not SK-Hynix-exclusive. Fires the step-4 falsifier on hbm-cowos-as-binding-bottleneck. From 2026-06-01-autoresearch-hbm-cowos-bottleneck-2026-corroboration.
  • Vera Rubin HBM4 certification (June 5, 2026): Jensen Huang certified Samsung for Vera Rubin HBM4 alongside SK Hynix and Micron on June 5, 2026. Samsung receives an estimated ~25-30% of Vera Rubin HBM4 volume (analyst estimates). Samsung began HBM4 mass production in February 2026, recovering from earlier qualification delays — the June 5 cert is Samsung's formal re-entry into the AI supply chain as a top-tier HBM4 supplier. Plans ~50% HBM capacity increase in 2026. Samsung memory chief Kim Jaejune stated "significant shortages will persist through at least 2027 and beyond." From 2026-06-08-autoresearch-hbm-supply-sk-hynix-samsung-micron-helium-2026.
  • Helium exposure: most exposed among major suppliers — Samsung buffer estimated at 6-12 weeks (shortest of the three HBM makers). Korea sourced 65% of its helium from Qatar in 2025; Taiwan 69% from Gulf sources. Samsung's shorter buffer relative to SK Hynix (8-16 weeks) and Micron (12-24 weeks) is a material near-term risk if South Ras Laffan does not restart before buffers are exhausted. Advanced AI chips and HBM receive triage priority, but Samsung's margin of safety is the thinnest. From 2026-06-08-autoresearch-hbm-supply-sk-hynix-samsung-micron-helium-2026.

Strengths (from a thesis-input perspective)

  • 70% SF2P yield is competitive — within striking distance of TSMC N2 (~65–75%)
  • Taylor fab's US location satisfies the same CHIPS-era political calculus as Intel's Arizona fabs
  • Qualcomm/AMD pipeline reduces Samsung's customer-acquisition risk independent of Apple decision
  • 4 years of GAA learning means fewer yield surprises than Intel's first-generation 18A

Weaknesses (from a thesis-input perspective)

  • No government champion at the scale of Trump White House brokering Apple-Intel
  • Intel has first-mover on the Apple preliminary deal; Intel's deal "puts Samsung under pressure" on Apple ambitions
  • Samsung foundry Q1 2026 operating loss ~KRW 1T; profitability recovery depends on 4nm recovery, not 2nm ramp
  • 2nm Taylor production is H2 2026 start for risk production; high-volume is 2027+ realistic
  • Korean HQ introduces geopolitical concentration risk of its own (Korea-China relationship)

Sources

Related

Referenced by
Mechanisms
HBM pre-committed + CoWoS half-owned by Nvidia → DRAM spot price +158% → memory/packaging bottleneck downstream of foundryJan 2026 tariffs + Taiwan trade deal + CHIPS Act expansion → 5pp cost-incentive shift → US-domiciled fab beneficiary stackSamsung SF2P yields hit 70% + Taylor 90% ready + Apple visit → genuine third foundry leg → order flow splits, not concentratesTSMC saturation → 4-year price hikes → marginal buyers shift → Intel anchor stack → re-rate
Concepts
Contract structure, not time, is the seam in the memory marketPicks-and-shovels: leading-edge fab buildoutTSMC capacity shortfall and 4-year pricing powerUS fab capacity bottleneck → Intel as the alternative foundry
Questions
Does Samsung have a credible leading-edge foundry comeback that would weaken the Intel-as-alternative thesis?Has Intel 18A reached yield parity with TSMC, or are partnerships outpacing the technical reality?
Entities
AppleASMLIntelSK HynixTSMC (Taiwan Semiconductor Manufacturing Co.)