TSMC’s annual report for the 2025 fiscal year, filed on Form 20-F, states that most of its production sits in Taiwan and names the outbreak of military conflict, war or hostilities as a risk to its business. We read that filing against TSMC’s own July earnings call and Samsung Electronics’ second-quarter release. The nearer story is not a war. It is a packaging queue TSMC says is already limiting its customers.
Last verified: 17 August 2026. Built from TSMC’s Form 20-F for FY2025 and its second-quarter 2026 earnings call, then Samsung Electronics’ second-quarter 2026 results release, in that order. The filed language holds until the next annual report. The capacity position moves every quarter, and four claims that circulate with this story are withheld below because we could not trace them to a body that published them under its own name.
Editor’s opinion
The Samsung foundry case does not need a Taiwan war, and that is also what makes it weak
Our view is that framing Samsung Electronics as a hedge against a Taiwan Strait conflict gets the mechanism backwards. A blockade or a strike is unforecastable, undateable and, for a portfolio, unhedgeable in any honest sense. What is dateable is that TSMC told analysts in July that its advanced packaging capacity had grown tight enough to constrain its customers, and said it would welcome others taking some of that load. Work moves for that reason, on ordinary quarterly timescales, whether or not anything happens in the strait.
The strongest point in favour is that both halves of the trade are now company-published rather than inferred. TSMC has put the concentration and the conflict risk in a signed filing, and Samsung has put expanding 2nm design wins for high-performance computing in its own results release. Neither claim depends on a broker note. A reader can open both documents.
The case against, at full strength
A packaging queue is not wafer allocation, and almost everyone repeating this story blurs the two. Welcoming flexibility in assembly says nothing about who fabricates the die, and TSMC has been adding packaging capacity of its own throughout. Samsung has disclosed design wins, not shipments, not volumes and not a customer name, and yield at volume settles foundry share, where the gap has been real for years. The view survives because it needs only marginal reallocation, not a change in leadership. It would not survive a Samsung yield disclosure showing the 2nm gap widening, and we would drop it on that.
Andrew Lee. This section argues a view rather than establishing a fact. It is not a forecast and not a recommendation, and the material it draws on is listed in the sources below.
TSMC writes the war risk into its own annual report
TSMC files an annual report on Form 20-F with the United States Securities and Exchange Commission, and the report for the 2025 fiscal year states that Taiwan holds most of the company’s production facilities. Its risk factors say that the outbreak of military conflict, war or hostilities could adversely affect its business and results of operations. That is not an outside assessment of TSMC’s exposure. It is TSMC describing its own, in a document the company signs and files.
A risk factor is not a forecast, and treating it as one is the most common error made with filings like this. It is closer to a lawyer’s inventory of everything that could go wrong, drafted to survive litigation rather than to predict anything. The same section of the same report lists earthquakes, power supply and customer concentration in the identical flat register. You should read it as evidence of what the regulator obliges the company to disclose, not as evidence of what the company expects to happen.
It is useful anyway, for a reason that has nothing to do with probability. Every large foundry customer reads that same page. A procurement team at a hyperscaler does not need TSMC to forecast a blockade; it needs an answer to a board question about depending on a single island, and the filed language is the answer it has to work with. Concentration becomes a purchasing agenda item long before it becomes an event, and purchasing agendas are what fund second sources.
The constraint that exists today is packaging, not conflict
The binding constraint right now is not on the wafer. It is on what happens after the wafer. On its second-quarter 2026 earnings call on 16 July, TSMC’s chief executive, C.C. Wei, told analysts that the company’s advanced packaging capacity had run tight enough to limit customers’ growth. In the same session he said he welcomed additional flexibility in the market from competing packaging technology, and that he hoped others would take some of TSMC’s load.
That is worth reading twice. The dominant supplier of the most contested manufacturing step in the industry said, on its own call, that it would like other people to take some of the work. Companies do not normally invite competitors in. They do it when the queue has become the reputational problem, and when the customers held up in that queue are the largest buyers they have.
The counter-fact belongs in the same breath. Packaging is not the leading-edge wafer, and the distinction carries the whole argument. A customer can move an assembly step without moving the die that goes into it, in the way a manufacturer can change who boxes a product without changing who builds it. Nothing in those remarks says TSMC is releasing wafer allocation.
We paraphrase these remarks rather than quote them. The call is TSMC’s own event and we know the date, but we do not hold an address for the transcript, and reconstructing an executive’s exact English wording out of a secondary Korean rendering would put quotation marks around a sentence we have not read. Where we could not verify the wording, we give the substance and say so.
Outside Taiwan the leading-edge list is short, and Samsung is on it
So who is in a position to take shared load at the leading edge? Samsung Electronics is the one company outside Taiwan that has told the market, in its own release rather than through reporting, that it is winning leading-edge foundry work now. In its second-quarter 2026 results announcement on 30 July, Samsung said customer design wins on its 2nm process for high-performance computing were expanding. The same release set out second-half plans: ramping the second generation of that 2nm process, and expanding sales of 4nm-based LPU and base-die products.
Those statements are narrow, and the narrowness is what makes them usable. A design win means a customer committed a chip to the process. It does not mean wafers shipped, it carries no volume, and Samsung named no customer. Yield at volume decides foundry position, and a results release settles none of that. We have also declined to print the industry’s standard line that Samsung is further ahead than Intel in serving external customers, because no named body publishes it.
If you want exposure to this business, the practical detail is that Samsung Electronics trades in Seoul and has no United States listing, so the ticker most foreign investors reach for first is not the company itself. We set out what that ticker actually is in our piece on SSNLF. The distinction matters more here than usual, because the foundry division is the part of Samsung this argument turns on, and no exchange lists it separately.
What we could not confirm
Five claims travel with this story that we are not printing. Three appear as rows below, and the sources list names the other two. Each one is specific, commentators repeat each with confidence, and none traces to a body that published it under its own name. Two of them are the figures that make the story exciting, which is the usual pattern.
Where each claim stands
Seven claims in this article
Where each one stands, as at 17 August 2026
| Claim | Attributed to | Status |
|---|---|---|
| Most production facilities located in Taiwan | TSMC, Form 20-F for FY2025, filed with the SEC | On the record, filed |
| Military conflict, war or hostilities named as a risk factor | TSMC, Form 20-F for FY2025, filed with the SEC | On the record, filed |
| Advanced packaging capacity limiting customer growth | TSMC second-quarter 2026 earnings call, 16 July 2026 | On the record, transcript address not held |
| 2nm HPC design wins expanding; second-generation 2nm and 4nm base-die plans | Samsung Global Newsroom, 30 July 2026 | On the record, company-published |
| Nvidia’s share of TSMC’s 2026 advanced packaging capacity | Brokerage estimates circulating in financial media | Not confirmed — figure withheld |
| Google moving CPU and TPU work to Samsung from 2028 | Trade reporting; neither company confirming | Not confirmed — claim withheld |
| Taylor target capacity, second-fab timing and total investment | Trade reporting; no newsroom text held | Not confirmed — figures withheld |
The table covers the claims we rely on as well as the ones we drop. One withheld item has no row because it is a claim about a claim: reporting widely describes TSMC’s June annual general meeting as having called advanced packaging capacity sold out for 2026, but that reporting names no outlet we can cite, and the July call already carries the same substance from the company directly. Every row marked withheld stays withheld until a named body publishes it. Nothing in this article depends on any of them.
What would change our view
The argument rests on one observable: whether leading-edge capacity outside Taiwan grows on a schedule somebody has committed to, rather than announced. We separated committed from announced for Samsung and SK hynix through 2028 in an earlier piece, and the same test applies here without modification. A Samsung newsroom release carrying the Taylor capacity and investment numbers would move that row out of the withheld list in a day. A named foundry customer disclosed by either party would do more than any capacity figure. Going the other way, a yield disclosure showing the 2nm gap widening would weaken the case regardless of how long TSMC’s packaging queue runs.
Nothing here forecasts a conflict in the Taiwan Strait, and we hold no view on whether one occurs or when. This article treats TSMC’s filed language about that risk as a fact about disclosure, which is what it is, and not as a probability, which it is not.
The market is pricing this reallocation off a queue you can date and a filing you can read, not off a scenario nobody can.
Sources
We take facts from these and write our own sentences. The lists below split by evidence status rather than by subject, because this article rests on two filed documents while trade reporting we could not trace to a named publisher surrounds it on every side. We link to a document only where we hold the exact address for it; where we do not, we give you enough to find it. The third list holds material we decline to reproduce, which is a statement about this article rather than a source for it.
Filings and company releases, read at source
- Taiwan Semiconductor Manufacturing Company, Form 20-F, fiscal year 2025, filed with the United States Securities and Exchange Commission. We take the geographic concentration of production and the armed-conflict risk factor from this document.
- Samsung Electronics, Samsung Electronics Announces Second Quarter 2026 Results, Samsung Global Newsroom, 30 July 2026. We take the 2nm high-performance-computing design wins, the second-generation 2nm ramp and the 4nm LPU and base-die plans from this release.
Named source, address not held
- Taiwan Semiconductor Manufacturing Company, second-quarter 2026 earnings conference call, 16 July 2026. Remarks by chief executive C.C. Wei on advanced packaging capacity and on welcoming additional flexibility from competing packaging technology. TSMC’s own event, on a fixed date, but we do not hold the transcript and therefore paraphrase rather than quote.
Material we decline to reproduce
- Brokerage estimates of one customer’s share of TSMC’s 2026 advanced packaging capacity, circulating in financial media. We can cite a research house estimate where the material names and dates the house; the versions reaching us do neither consistently, so we withhold the figure.
- Trade reporting on a prospective shift of Google CPU and TPU production to Samsung foundry from 2028. Neither company has confirmed it in any document we hold, so we withhold the claim.
- Trade reporting on Samsung’s Taylor, Texas target wafer capacity, second-fab construction timing and total investment. We obtained no Samsung newsroom text carrying these figures, so we withhold them.
- Reporting of TSMC’s June 2026 annual general meeting describing advanced packaging capacity as sold out for 2026. We identified no citable outlet, and the July earnings call carries the same substance from the company, so we do not rely on the meeting.
- The general industry assessment that Samsung is further ahead than Intel in serving external foundry customers. Widely repeated, published by no named body, and therefore not stated as fact in this article.
Corrections
None yet. If you can point to a primary document that contradicts anything above, we will correct it and say what changed.
Disclosure. The author holds no position in Samsung Electronics, TSMC, Intel or Nvidia as of the date of publication. This article is general information and analysis about semiconductor foundry capacity and the geopolitical risk that chip manufacturers disclose in their filings. It is not investment advice, not a recommendation to buy or sell any security, and not a forecast of any price or index level. The section headed Editor’s opinion is comment: it argues a view, and we fence it off from the sourced material for that reason. We have not tailored anything here to your circumstances, and only a qualified adviser in your jurisdiction can do that. We are not registered as an investment adviser in any jurisdiction, and this site has no commercial relationship with any company, exchange or research provider named above. See the Editorial & Sourcing Policy and Disclaimer.