brain/
sourcestock-market

Autoresearch: Samsung foundry Q2 as a third alternative to Intel/TSMC

Post-May-27 Q2 2026 foundry check: DS profit still HBM; 2nm yield ~60% vs 70% bar; Qualcomm flagship on TSMC; Broadcom $200B MOU; Taylor Fab 2 2030; High-NA slipped to 1nm ~2030

Source

Autoresearch: Samsung foundry Q2 as a third alternative to Intel/TSMC

Generated by /autoresearch on 2026-08-24. Synthesized across 3 rounds from 11 fetched pages (plus 1 earnings-call transcript extracted via search dump), anchored by Samsung Electronics. See Provenance. Treat as raw material — review before promoting into a project or thread. Context: vault/projects/stock-market Prompt: the open question on samsung-foundry-as-third-alternative — Does Samsung have a credible leading-edge foundry comeback that would weaken the Intel-as-alternative thesis? Scope is the dated miss after the last research source (2026-05-27): Q2 2026 foundry vs HBM mix, yield, Qualcomm/AMD/Taylor, Apple deal acknowledgement or repudiation, and 2027 High-NA allocation. Not a re-run of the May partnership tape.

Summary

Samsung's Q2 2026 print is a memory event, not a foundry-profit event. Device Solutions posted KRW 127.5 trillion revenue and KRW 89.2 trillion operating profit; official foundry language is "improved significantly prior to incentive-related provisions," with HBM base-die demand and U.S. orders as the drivers (Samsung Semiconductor Newsroom Q2 2026). Trade recaps put memory at 99.7% of group operating profit and keep foundry in the red (TrendForce, 2026-08-03). On the question's load-bearing tests: 2nm yield is now reported around 60% (up from the April ~55% the May 27 source used), still below the ~70% bar Qualcomm and AMD are said to need (TrendForce, 2026-08-03; TrendForce, 2026-08-21). Qualcomm's next flagship is reported exclusive to TSMC N2P (WCCFTech). AMD's official Venice ramp is TSMC 2nm, not Samsung (AMD IR, 2026-05-21). No fetched post-May-27 page acknowledges or repudiates a public Apple–Samsung leading-edge SoC deal. The new customer signal is a July 25 Broadcom MOU the companies "expect" at more than $200 billion across memory and foundry through 2030 — a framework, not a disclosed wafer contract (Samsung Semiconductor Newsroom, 2026-07-25). High-NA is not a 2027 foundry-scale allocation: commercial High-NA is delayed to the 1nm-class node around 2030 (SamMobile / NGL 2026). Bottom line for the thesis: Samsung is a more credible 4nm / HBM-base-die / utilization third, and has a named 2nm HPC framework customer. It is not yet a co-equal leading-edge alternative that would split Apple / Qualcomm flagship / AMD Venice volume away from the Intel-as-alternative story.

Findings

Q2 2026 is HBM-driven; foundry improved pre-provision, still not a disclosed profit center

Samsung's official Q2 2026 release (quarter ended June 30, 2026): consolidated revenue KRW 171.5 trillion (+28% QoQ); operating profit KRW 89.5 trillion; DS KRW 127.5 trillion revenue and KRW 89.2 trillion operating profit. Memory set another quarterly revenue and profit record. Foundry: "Earnings for the Foundry Business improved significantly prior to incentive-related provisions, driven by HBM base-die demand and strong orders from customers in the U.S. Design wins with major customers continued to expand, with 2nm HPC engagements as a notable example." H2 plan: ramp "new mobile products based on the second-generation 2nm process" and expand "LPU and base-die products based on the 4nm process"; double-digit foundry revenue growth targeted (Samsung Semiconductor Newsroom Q2 2026).

The July 29, 2026 earnings call (Roic AI transcript dump) matches the release and adds color the newsroom omitted. CFO Soon-Cheol Park: Foundry "gained from higher utilization rates and growing demand for our advanced node while continuing to expand orders from major customers for 2 nano HPC." Foundry sales head Sukchae Kang: Q2 revenue rose on "Memory HBM-based type products as well as product demand centered on customers in the U.S."; earnings "showed significant improvement before incentive-related charges." H2: mass production of "new mobile-oriented products based on the second-generation 2-nano process," volume LPU on 4-nano, and more memory base-die sales. Advanced-node revenue "projected to exceed 50%"; AI/HPC mix "from the high-teens percentage in 2025 to over 30% in 2026." On turnaround timing: "it is difficult to predict exactly when we will turn around due to the nature of the foundry business, which is driven by customer orders"; "the turnaround is possible in the near term." Utilization: "improved year-over-year across all nodes. Advanced nodes, 8-nano and below, actually have reached full utilization." On yield, Kang gave no percentage: "building on the success for mass production of first gen 2-nano technology, we will be ramping up second-generation 2-nano production in the second half" and "expect to be able to stabilize yields." Taylor CapEx "increased quarter-on-quarter" and "remains on track" (Roic AI Q2 FY2026 transcript).

TrendForce's August 3 recap of Korean press: memory contributed 99.7% of total operating profit; foundry "remained in the red"; Newsis estimated Foundry + System LSI combined loss around KRW 600 billion in Q2; Chosun Biz says foundry aims at 100% utilization in 2H from a current 70–80% range; HuffPost Korea cited an analyst projecting a possible Q3 foundry profit on HBM4 base dies plus Exynos (TrendForce, 2026-08-03). Seoul Economic Daily, writing off the same call: Q2 foundry revenue rose on 4nm HBM base dies and a U.S. customer; "performance on a pre-provision basis also improved significantly"; "price increases and yield improvements overlapped with rising utilization" (Seoul Economic Daily, 2026-07-30).

Implication vs May 27: the Q2 profit-path claim the question was watching is not a disclosed foundry operating profit. DS's KRW 89.2 trillion is HBM. Foundry is fuller and less loss-making on a pre-provision basis; standalone foundry P&L is still unpublished.

2nm yield: ~55% in April, ~60% by August — still below the ~70% customer bar

Samsung did not disclose a 2nm yield percentage on the Q2 call (Roic AI transcript). The trade-press series that the May 27 source treated as authoritative has moved the number up, not through the bar:

  • TrendForce July 17, citing Chosun Ilbo: 2nm yield "reportedly improved to around 55%, approaching the roughly 60% threshold generally regarded as necessary for stable mass production" (TrendForce, 2026-07-17).
  • TrendForce August 3, citing Chosun Biz: Samsung "needs to raise 2nm yields to around 70% to secure large-scale mass production contracts, with current yields estimated at roughly 60%" (TrendForce, 2026-08-03).
  • TrendForce August 21, citing Seoul Economy News: industry watching whether Samsung can lift 2nm yields "from an estimated 60% to the 70% range — a level considered necessary to secure large-scale production from major customers such as Qualcomm and AMD"; SF2P "expected to enter mass production for mobile products in the second half of 2026" (TrendForce, 2026-08-21).

WCCFTech, citing Business Korea and rating the rumor 55% "plausible": Snapdragon 8 Elite Gen 6 / Gen 6 Pro "expected to be exclusively mass produced on TSMC’s 2nm ‘N2P’"; Samsung 2nm GAA "haven’t surpassed the 60 percent mark"; Qualcomm's required benchmark is "70 percent"; dual-source remains the hoped-for path if yields clear (WCCFTech).

Contradiction to record: the May 27 source treated April TrendForce ~55% as the production average and the January "70% SF2P" print as a test-lot. August Chosun/TrendForce now say ~60%, still below 70%. That is a modest improvement, not a walk-back of the April figure and not a crossing of the mass-production bar the question uses. Official Samsung still refuses a number.

Qualcomm flagship still TSMC; Qualcomm is already an SF4 customer

The fetched Qualcomm evidence is not a closed 2nm flagship dual-source. WCCFTech's Business Korea retell says Gen 6 / Gen 6 Pro exclusive to TSMC N2P because Samsung has not hit Qualcomm's 70% quality bar (WCCFTech). TrendForce August 21, citing Reuters, says Qualcomm is already a customer on SF4, and Google is "in talks" for SF4, while NVIDIA's Groq-3 inference LPU is lifting 4nm demand; Samsung's 4nm lines are also full making its own HBM4 base dies (TrendForce, 2026-08-21).

That splits the Qualcomm question: 4nm / mid-tier relationship exists; 2nm flagship still TSMC. The May "Qualcomm still mulling" line has not become a 2nm flagship win.

AMD Venice is official TSMC 2nm; no fetched Samsung close

AMD's May 21, 2026 press release: EPYC "Venice" "is ramping production in Taiwan on TSMC’s advanced 2nm process technology, with future plans to ramp production at TSMC’s Arizona fabrication facility." Lisa Su and TSMC's C.C. Wei both speak only to the TSMC partnership. The follow-on "Verano" is also framed as TSMC 2nm (AMD IR, 2026-05-21).

This pass fetched no post-May-27 AMD or Samsung primary that confirms a closed Samsung 2nm CPU order. The May "AMD evaluating SF2P for Venice" line from the existing question page is not corroborated by AMD's own Venice announcement — Venice production is disclosed as TSMC. Any residual Samsung AMD volume, if it exists, is not in the fetched record.

Broadcom MOU is the new named 2nm customer — framework, not a wafer contract

On July 25, 2026, Samsung and Broadcom announced an MOU at an AI Summit in San Francisco (Jinman Han; Hock Tan; Korean government representatives). "The companies expect the collaboration to be estimated at more than $200B across memory and foundry over the next five years through 2030." Memory: HBM for Broadcom's next-generation AI accelerators. Foundry: "Samsung’s 2-nanometer (nm) and below process technologies for Broadcom’s products, including Wireless Broadband Communications (WBC) solutions," plus 2.3D / 2.5D packaging on 2nm (Samsung Semiconductor Newsroom, 2026-07-25).

Kang on the Q2 call, five days later: "We're also in talks with Broadcom and other major customers on diverse other projects"; 2-nano project wins expected to "increase by more than double year-over-year in 2026"; "secured 2-nano projects from major CSPs and AI/HPC customers" already in "the project design phase" (Roic AI transcript).

Read this as a named 2nm HPC framework, not as $200 billion of booked foundry revenue. The newsroom wording is "expect" / "estimated" / "MOU." It does change the May picture: the largest public post-May customer move is Broadcom + unnamed CSPs in design, not Qualcomm or AMD flagship closes.

Apple leading-edge SoC: no fetched acknowledgement or repudiation after May 27

The question asked for a public Apple–Samsung deal acknowledgement or repudiation. This pass fetched no such page. Web searches for July/August 2026 Apple–Samsung foundry orders returned (a) May 2026 Bloomberg-retell exploratory talks (Apple executives at Taylor; no orders) and (b) August 2025 Austin image-sensor / CIS stories that are not a 2026 A-series / M-series foundry award. Those search hits were not fetched and are not cited as evidence. The Apple test on the question page remains open.

Taylor: Fab 1 in 2026, Fab 2 groundbreaking late 2026 / MP 2030

Kang on the call: "Taylor Fab 1 is preparing to commence operations by 2026 as planned, with plans to gradually ramp up 2-nano capacity." "Taylor Fab 2 will begin construction by the end of this year, targeting mass production by 2030." "We're seeing an increase in customer inquiries for 1.4-nano as well" and reviewing "additional fab capacity," with details "in phases based on customer engagements" (Roic AI transcript). TrendForce (Yonhap / JoongAng recap) and Seoul Economic Daily repeat the same call points: Fab 1 2026, Fab 2 groundbreaking end-2026 / MP 2030, 1.4nm space under review, 2nm orders expected to more than double YoY, 8nm-and-below full (TrendForce, 2026-07-31; Seoul Economic Daily, 2026-07-30).

U.S. leading-edge capacity that could take Apple / CSP volume is therefore 2026 start / 2027 ramp for Fab 1, not a 2026 volume alternative to Intel or TSMC Arizona. Fab 2 is a 2030 story.

High-NA: two tools installed; volume production slipped to 1nm ~2030

The May question asked whether Samsung scales High-NA adoption for 2027. Fetched evidence says no.

TrendForce July 17, citing Chosun Ilbo: Intel is first to put High-NA into mass production on selected 18A Panther Lake layers; Samsung has installed two High-NA systems (Hwaseong last year; second in H1 2026; just over KRW 1 trillion) but "has yet to officially confirm that the systems have entered mass production." The hold is framed as cost / depreciation while foundry is still loss-making, not as a tool-allocation loss. Same piece still had 2nm yield "around 55%" (TrendForce, 2026-07-17).

SamMobile's write-up of NGL 2026: Park Chang-min (Foundry Process Development) said Samsung will delay full commercial High-NA to the 1nm-class (A10) node, "expected around 2030." 2nm and 1.4nm stay on 0.33 NA EUV + multi-patterning because the High-NA ecosystem (masks, pellicles, materials) is immature and each tool is "$350 million to $400 million." Intel is already using High-NA on selected 18A layers; TSMC is mostly R&D, commercial High-NA around 2029–2030 (SamMobile).

Implication vs May: the May "Samsung 2 High-NA units in 2026 — equal to Intel's allocation" line is still true as tool count. It is not true as 2027 High-NA volume leadership. Intel is using High-NA in production; Samsung is not, by its own NGL statement.

Pricing power showed up on 4nm/5nm/8nm, not as a 2nm yield victory

TrendForce August 21, citing Reuters and ijiwei: Samsung raised 4nm-led advanced-node prices by up to 15% (China/U.S. SF4 +10–15%; Taiwan +5–10%); SF5 +10–15%; 8nm ~+10%. Context is tight capacity (Chosun 70–80% heading to full in 2H), HBM4 base-die fill on SF4, Groq-3 LPU, and Google SF4 talks — not a claim that 2nm has reached TSMC-like yields. A year earlier Samsung was still discounting 2nm (Tom’s Hardware via TrendForce: ~$20,000/wafer vs TSMC ~$30,000) to win Tesla AI6 (TrendForce, 2026-08-21).

This is utilization-and-scarcity pricing on nodes Samsung can already run full. It does not by itself make SF2 a third leading-edge alternative.

Contradictions and open questions

  • 2nm yield 55% vs 60% vs January 70%. April TrendForce ~55% (the May 27 source of record) vs July Chosun ~55% vs August Chosun/TrendForce ~60%. Official Samsung: no number. Do not silently retire the April 55% figure; treat August 60% as the new trade-press estimate, still below 70%. The January 70% SF2P claim remains unconfirmed as a production average.
  • Utilization: official "full" on 8nm-and-below vs Chosun 70–80% overall. Both can be true (advanced full, blended 70–80%). Do not collapse them into a single foundry utilization number.
  • Foundry profitability. Official: improved pre-provision; turnaround "near term." TrendForce/Newsis: still in the red; ~KRW 600 billion combined Foundry+LSI Q2 loss (estimate). No fetched Samsung segment P&L for foundry alone.
  • Apple leading-edge SoC. Still no fetched acknowledgement or repudiation after May 27. Exploratory Taylor visit remains the last cited wiki fact; this pass does not add a deal.
  • AMD. Official Venice is TSMC. May "evaluating Samsung" is not closed in any fetched post-May primary.
  • Broadcom $200B. Official MOU / "expect" / "estimated." Volume, node mix, binding offtake, and how much is HBM vs foundry are undisclosed.
  • 2027 High-NA allocation. No fetched ASML 2027 customer-split table in this pass. What was fetched is Samsung delaying High-NA volume to 1nm ~2030 while Intel is already in production on selected 18A layers.

Provenance

Rounds run: 3 of 3

Sub-questions by round:

Round 1 (broad survey):

  1. What did Samsung report for Q2 2026 foundry vs HBM (utilization, profit, customer revenue)?
  2. Post-May 2nm SF2 / SF2P yield — did they clear ~70%?
  3. Qualcomm / AMD 2nm deal status after May 27
  4. Apple–Samsung foundry: public deal, repudiation, or still Intel-led?
  5. Taylor TX and 2027 High-NA allocation updates

Round 2 (drill-down):

  1. Official Broadcom collaboration — targeting the July 25 newsroom item and Q2 "in talks with Broadcom" line
  2. AMD Venice primary — targeting whether the May "AMD evaluating SF2P" claim survived AMD's own announcement
  3. High-NA / Taylor confirmation from call recaps

Round 3 (resolve remaining uncertainty):

  1. High-NA volume-production timing (NGL 2026 / cost hold) — targeting the 2027 allocation test
  2. Taylor Fab 2 / 1.4nm from a second trade recap
  3. Search for any post-May Apple–Samsung leading-edge SoC acknowledgement (none fetched)

Anchor source (Grokipedia, fetched before round 1):

  • Samsung Electronics — 20,017 chars extracted via grokipedia-fetch helper — primer on foundry share, GAA, Taylor delay history, and 2026 profitability-target language. Fast-moving Q2 2026 figures in this clipping come from primaries and trade press, not from the encyclopedia page.

URLs fetched (11 successful page fetches, 1 search-dump transcript, 1 failed):

Round 1:

Round 2:

Round 3:

Not fetched (search only; not cited as evidence): Reuters, Bloomberg, Digitimes, Tom's Hardware, eWeek, Investing.com, Korea Herald, Businesskorea August 2025 CIS stories, KeyBanc Apple-Intel notes.

Tools used: WebSearch, WebFetch, grokipedia-fetch (skill / _lib/grokipedia.py). Generated: 2026-08-24 12:45 UTC

Referenced by
Questions
Does Samsung have a credible leading-edge foundry comeback that would weaken the Intel-as-alternative thesis?
Entities
Advanced Micro Devices (AMD)BroadcomSamsung Foundry