That distinction matters. A leading-edge fab is not capacity merely because it has been announced, funded, or even fitted with a corporate label. It must move through construction, completion, tool installation, qualification, and production ramp-up. The evidence available here confirms substantial expansion and meaningful labor pressure, but it does not prove delays at any particular site or quantify the eventual impact on wafer output. It also does not establish that labor is a more important constraint than capital overall.
For Samsung Electronics, the development is best read as a reminder that foundry competition depends on execution across manufacturing, customer engagement, and workforce availability—not as proof that TSMC’s construction challenge will automatically redirect business to Samsung. For Windows PC buyers, the near-term implication is even more qualified: the risk is continued tightness around the most advanced silicon and packaging, rather than a verified prediction of a specific laptop, GPU, or CPU shortage.
First, untangle the fab counts
The dramatic claims about TSMC building “20 fabs” or “25 facilities” refer to different scopes and should not be used interchangeably.
At the September 2 SEMICON Taiwan event, TSMC executive Hou Yongqing was reported as saying that the company was simultaneously advancing 13 wafer fabs in Taiwan and five to six overseas. That produces the widely reported total of nearly 20 wafer fabs globally. He also described the company’s expansion pace as roughly four to five times its previous rate, while saying it was still unable to keep up with demand.
A separate count of 25 is broader. It explicitly includes both chip-manufacturing fabs and advanced-packaging facilities worldwide. Advanced packaging is strategically important, especially for AI processors that need to combine compute dies and high-bandwidth memory, but it is not the same thing as a wafer fab. Treating the 25-facility number as 25 wafer fabs overstates what that figure means.
There is another timing distinction. In its July 2026 earnings conference, TSMC said it was building 13 leading-edge and advanced-packaging fabs in Taiwan over the next several years. That is a multiyear plan and cannot, on its own, establish that every one of those facilities was an active construction site at that moment.
Arizona provides the clearest illustration of why project-stage language matters. TSMC’s first Arizona fab has been in N4 volume production since late 2024. Construction on the second fab was complete, with volume production expected in 2027. The company also announced an additional US$100 billion commitment for Arizona, intended for several more 2nm-and-below logic fabs and advanced-packaging fabs. Those additional plans brought its announced Arizona investment total to US$265 billion, covering 10 fabs, two advanced-packaging facilities, and an R&D center.
In other words, a regional footprint can contain a fab already shipping production, another that is structurally complete but not yet in volume output, and later facilities that remain future projects. Announced investment is important, but it is not a synonym for immediately available semiconductor supply.
Construction labor is an immediate constraint—not a complete explanation
Hou’s reported warning was specific: construction-worker shortages are an immediate limitation, affecting TSMC’s Taiwan operations and overseas sites including Arizona. That is credible as a practical constraint because fab projects require a large local workforce and a sequence of specialized construction work that cannot simply be shifted to a different country or completed purely by spending more money.
Arizona’s regional numbers underscore the scale of the workforce challenge. A local report cited a projection of a shortage of roughly 10,000 workers per year in the state, while an academic estimate suggested that TSMC’s additional US$100 billion could generate 10,000 to 12,000 construction jobs. These figures support the conclusion that labor demand will be intense. They do not prove that a TSMC fab has slipped its schedule, nor do they prove a direct effect on finished-chip output.
It is tempting to frame this as a simple “money cannot solve it” story. The available record does not support that certainty. TSMC increased its 2026 capital-expenditure guidance to US$60 billion to US$64 billion, up from US$52 billion to US$56 billion, while also making the new Arizona commitment. That demonstrates access to, and willingness to deploy, very substantial capital.
But the evidence does not compare the binding effects of financing, skilled construction labor, equipment availability, permits, utilities, supply chains, or later manufacturing ramp-up. Labor has been identified as an immediate problem. It has not been proven to be the sole bottleneck or the overriding constraint across all sites.
That nuance is important for technology readers. The semiconductor industry often reports capacity in enormous dollar figures and multiyear fab totals, while users understandably want to know when a product will become easier to buy. The path between those two facts is not automatic. Construction is only one phase; completing a building is not the same as qualifying tools, achieving yields, or shipping sufficient volumes to customers.
Why advanced packaging still deserves attention
TSMC’s expansion plans include advanced packaging, not just logic-wafer manufacturing. This deserves careful attention because the performance and availability of many AI-oriented systems depend on more than the leading-edge compute die. Packaging capacity can matter to the final availability of a complete accelerator or processor.
However, it would go too far to turn that general reality into a prediction that particular TSMC customers will not receive near-term supply relief. The reviewed material does not disclose customer-specific allocations, node-level capacity, packaging output, or a timetable for supply conditions to improve. It therefore cannot verify claims about the prospects for Nvidia, AMD, or any other named customer.
For Windows users, the reasonable conclusion is modest. Continued strain in leading-edge manufacturing and advanced packaging can remain an indirect risk for products that rely on the newest processors and AI accelerators. That could affect the pace at which highly capable Windows workstations, gaming hardware, and AI PCs reach broad availability or lower price points. But the evidence does not support forecasting a shortage of a particular Windows device, a specific GPU, or a defined date for market relief.
The most useful buying takeaway is to separate need from speculation. Buyers with an immediate workstation or gaming-PC requirement should evaluate the hardware available today rather than assume a future expansion announcement will quickly transform pricing or supply. Organizations planning AI infrastructure should likewise treat fab announcements as a longer-term capacity signal, not as a confirmed delivery schedule for their preferred chips.
Samsung has a potential opening, but no demonstrated windfall
TSMC’s physical expansion challenge naturally raises the question of whether Samsung Foundry could benefit. Samsung’s own second-quarter 2026 results show meaningful activity: it said foundry earnings had improved significantly before incentive-related provisions, cited expanding design wins with major customers, and highlighted 2nm high-performance-computing engagements. For the second half of 2026, Samsung said it planned to ramp mobile production using its second-generation 2nm process and was targeting double-digit foundry revenue growth.
Those disclosures establish that Samsung is trying to advance its foundry position, including at 2nm. They do not establish that Samsung has received orders because of TSMC’s construction-worker shortage. Nor do they show that Samsung can absorb any defined amount of displaced demand, offer equivalent capacity at the same time, or match a customer’s packaging and manufacturing requirements.
That last point matters because a chip design is not always readily portable between foundries. Process technology, libraries, design rules, packaging choices, production qualification, and customer risk tolerance can all shape whether a second source is practical. The dossier does not provide evidence about any particular customer’s portability plans, so a direct order-transfer narrative would be speculation.
There is also a claim circulating that Samsung is mobilizing company-wide resources through engineering, procurement, construction, tool hook-up, commissioning, and yield stabilization. The evidence reviewed does not independently substantiate that detailed assertion. It was attributed to an unnamed construction executive rather than to a verifiable Samsung disclosure. It should not be presented as established fact.
What Samsung has disclosed is more concrete, if less dramatic. At its Taylor, Texas campus, Samsung says it expects to create more than 1,800 direct high-tech jobs over the next decade and is working with workforce-development organizations, colleges, and universities to build a talent pipeline. That does not eliminate regional labor competition, but it shows Samsung recognizes workforce development as part of establishing a U.S. manufacturing base.
The U.S. manufacturing lesson: capacity is regional
Arizona and Texas are often discussed as though they are isolated corporate campuses. In practice, major fab clusters compete for overlapping pools of construction workers, technical specialists, suppliers, infrastructure, education partners, and local services. A company can have the balance sheet to announce billions in investment and still encounter bottlenecks in the surrounding labor market.
This is a public-policy issue as much as a corporate one. The reported Arizona workforce shortfall and job estimates suggest that local training pipelines, trade capacity, housing, transport, permitting, and infrastructure coordination can influence the practical pace of semiconductor expansion. Those measures do not guarantee a fab will open on time, but they affect the conditions under which a project can move from an investment pledge to an operating facility.
For policymakers, the lesson is not simply to count announced fabs or add up headline investment totals. Success should be measured by whether projects can recruit and retain workers, complete construction safely, bring equipment online, qualify production, and sustain operations. Each stage requires different skills and can face different constraints.
For the semiconductor industry, the same principle applies internationally. TSMC’s situation shows that locating facilities across Taiwan and the United States can diversify geography, but it does not remove the need for sufficient local labor at each site. Expansion spreads opportunity and risk; it does not make the labor requirement disappear.
What can be said with confidence
TSMC is pursuing an exceptionally large expansion, with nearly 20 wafer fabs reportedly moving forward worldwide under one count and 25 facilities under a broader count that includes advanced packaging. It has materially raised 2026 capital-spending guidance and expanded its announced Arizona commitment. Its leadership has also identified construction-worker shortages in Taiwan and Arizona as an immediate constraint.
Samsung, meanwhile, has disclosed 2nm HPC engagements, a planned second-generation 2nm mobile-production ramp in the second half of 2026, and workforce-development efforts connected to Taylor. Those are real indicators of competitive intent and operational preparation.
What remains unproven is the leap from these facts to a near-term shift in foundry leadership, customer orders, chip availability, or Windows hardware prices. The evidence supports a picture of sustained supply-chain pressure and difficult physical execution. It does not yet show which company will convert that pressure into the largest competitive advantage—or when consumers will see the results on store shelves.
That is the key distinction to keep in view: semiconductor investment is accelerating, but capacity is ultimately delivered by people, projects, production ramp-up, and time.