At its third-quarter 2026 (July–September) earnings call on October 29, 2026, Samsung Electronics ("Samsung") said utilization at its advanced process nodes of 8nm and below will exceed 100% in 2027. The company expects demand for AI semiconductors to outstrip its production capacity, and it projects that the number of 2nm projects it wins in 2026 will more than double from the previous year.
For Samsung, which is under pressure to improve the profitability of its foundry (contract chip manufacturing) business, the announcement could mark an important turning point. Production of advanced chips for AI accelerators and cloud infrastructure remains concentrated at TSMC, but Samsung's advanced nodes could become an option for cloud providers looking to diversify their suppliers. Samsung said it is indeed seeing more 2nm orders from customers, including major hyperscalers.
However, the figures presented are not actual production results or confirmed revenue. The utilization figure above 100% is a forecast for 2027, and a doubling in the number of projects does not mean wafer shipments or revenue will grow by the same proportion. Before ordered chips move from design to prototyping, validation and mass production, several challenges remain, including yield and manufacturing cost. We compare Samsung's disclosures with the movements of outside customers and TSMC's results to examine how far the foundry recovery has actually progressed.
What "Utilization Above 100%" and "2nm Projects More Than Doubling" Mean
At the October 29 earnings call, Kang Suk-chae, the executive vice president in charge of Samsung's foundry business, said demand across processes of 8nm and below would far exceed production capacity in 2027, pushing utilization above 100%. Korean outlet Financial News also reported the remarks. To meet rising demand, the company plans to expand capacity across its 8nm, 4nm and 2nm processes.
It is worth being careful about what "utilization above 100%" means. Fab utilization normally expresses how far actual or planned output reaches against a baseline, such as the number of wafers that can be processed in a given period. Samsung did not explain the baseline used to calculate 100%, or how much planned capacity expansion is factored in. The figure is therefore best read not as equipment running beyond its physical limits, but as an outlook in which demand and production plans exceed the baseline capacity.
The forecast also does not reflect third-quarter 2026 results. Even if the order outlook is strong, the number of wafers that can actually be shipped depends on line expansion, equipment installation and yield in mass production. A forecast that demand will exceed capacity alone does not show that the fabs can run at full load while securing a profit.
Another key metric is the outlook that 2nm project wins in 2026 will more than double year on year. Samsung said that, following the second quarter, it won 2nm projects in the third quarter from major cloud service providers (CSPs) and AI and high-performance computing (HPC) customers. It intends to stabilize its order base through long-term agreements (LTAs) with key customers.
The unit of this metric, however, is strictly the "number of projects." It does not mean that wafer starts, contract value or the number of customer companies will double. The actual project counts for 2025 and 2026 were not disclosed either, so it is difficult to assess the scale of orders in quantitative terms.
In contract chip manufacturing, winning a design project with a particular process does not mean mass production starts immediately. Customers must complete circuit and physical design, go through tape-out and photomask production, produce prototype wafers, and verify electrical characteristics and reliability. If design changes arise along the way, or the customer revisits its product plans, the originally planned production timing and order volumes can change. A rise in project numbers is an important leading indicator of future revenue growth, but an increase in sales or profit is not assured.
Taiwan's Faraday Wins a 2nm AI Chip Project, a Concrete Sign of Progress
There is evidence in the external supply chain that supports the expansion in advanced-node orders Samsung describes. On October 27, 2026, Taiwanese ASIC design services company Faraday Technology announced it had won a development project for a cloud AI chip using a 2nm GAA (Gate-All-Around) process from an overseas customer. It is the company's first 2nm GAA project and its largest ever in terms of design complexity, NRE (Non-Recurring Engineering) contract value, which covers upfront development costs, and the scale of advanced packaging technology involved.
According to a report by Taiwan's Central News Agency (CNA), Faraday President Wang Kuo-yung said at the earnings call that the project will use Samsung's foundry and that the company will also work with UMC and a memory maker. The total contract value reaches tens of millions of dollars, and Faraday expects to book its first revenue from the first half of 2027. This is consistent with the expansion in 2nm projects for cloud providers that Samsung announced.
How much of Samsung's production capacity the Faraday project will use is still unknown. Faraday's public earnings materials do not name Samsung, but CNA reported that company executives disclosed the manufacturing partner at the earnings call. That sets it apart from observations based on anonymous information, but wafer order volumes and the start date of mass production have not been confirmed.
In contract ASIC development, revenue from NRE contracts for design and prototyping must be considered separately from the revenue from wafer manufacturing and shipment that begins once a product enters mass production. The revenue Faraday said it would book in the first half of 2027 does not mean 2nm chip mass production will begin at the same time. Continuous mass production follows only after the customer's design is complete, prototype chips pass validation, and sufficient yield is secured.
Comparing Samsung's order expansion with TSMC's advanced chip business shows important differences in the figures the two companies disclose.
| Item | Samsung's disclosure | TSMC's results | Caveat for comparison |
|---|---|---|---|
| Advanced-node utilization | Forecast above 100% for 8nm and below in 2027 | No directly comparable figure | Samsung's is a forecast, and details of the calculation baseline are undisclosed |
| 2nm business | 2026 project wins expected to more than double year on year | 8% of wafer revenue in Q3 2026 | Project count and actual revenue share are different metrics |
| HPC share of revenue | Expected to exceed 50% in 2027 | 68% of net revenue in Q3 2026 | Samsung's is a forward-looking outlook; TSMC's is actual results |
| 2nm projects for external customers | Orders growing from major CSPs and AI/HPC customers | 2nm is already contributing to revenue | Samsung has not disclosed shipment volumes by project |
TSMC's figures are based on its Q3 2026 earnings materials. Samsung's figures are outlooks presented at the October 29 earnings call. Because what is measured and the periods covered differ, they should not be used for a simple head-to-head ranking.
The Faraday project is important as a concrete example showing that Samsung is starting to win external customers for advanced AI chips. Whether the ordered chips complete the design stage and lead to continuous shipments of production wafers will need to be confirmed as the work progresses.
Second-Generation 2nm and 4nm NPU Mass Production Will Shape Near-Term Growth
Ahead of the 2027 capacity expansion, Samsung expects double-digit growth in foundry revenue in the fourth quarter of 2026 (October–December), both quarter on quarter and year on year. In its official earnings release on October 29, the company said third-quarter profitability improved on higher sales at advanced nodes and rising demand for HBM base dies. However, because foundry-only revenue and operating profit are not disclosed, the actual amount of revenue growth or profit improvement cannot be calculated from the growth-rate outlook.
Three things are said to support fourth-quarter growth: the start of mass production of new mobile products using the second-generation 2nm process, expanded mass production of 4nm NPUs (neural processing units) for a major customer, and an increase in contract manufacturing for memory, including HBM-related products. Samsung also plans to grow sales to major customers in the US and China, improving the share of advanced-node revenue and fab utilization.
Korea's Seoul Economic Daily identifies the mobile product using the second-generation 2nm process as the "Exynos 2700." Samsung's official earnings release, however, does not name a specific product. Mobile SoCs designed by the company's System LSI division are key products for the foundry division when ramping an advanced process, but for this mass production plan too, the product name and the actual shipment status need to be confirmed separately.
The state of mass production differs by process. For 4nm, expanded production of NPUs for a major customer is planned, while the second-generation 2nm mobile product is at the stage of a full-scale production ramp. For 2nm products for a major external customer, Samsung plans to begin mass production of new products from the second half of 2027. In other words, Samsung's 2027 growth outlook contains two different challenges: expanding output on existing processes and ramping a next-generation process.
Especially important is the yield of the 2nm process. Samsung uses a GAA structure in which the transistor gate surrounds the channel on all sides, a technology the company calls MBCFET. While this improves electrical control of the channel compared with conventional FinFETs, forming nanosheets and processing the stacked structure demand sophisticated process control. At advanced nodes, if tiny defects reduce the share of working chips, the manufacturing cost per wafer can rise sharply.
Samsung has indicated it will work to stabilize 2nm yield, but the earnings call did not disclose specific yield or defect density figures. An increase in the number of 2nm projects or an announcement of the start of mass production alone therefore cannot establish that manufacturing costs have fallen sufficiently. A key test going forward is whether mass production of second-generation 2nm products, starting in the fourth quarter of 2026, stabilizes and whether that experience can be applied to the products for external customers in the second half of 2027.
Taylor Fab 1 to Start Operating in 2027; Fab 2 Planned to Break Ground in 2026
Samsung's semiconductor plant in Taylor, Texas, plays an important role in its advanced-node expansion. For major US technology companies, securing domestic advanced chip manufacturing capacity is also meaningful from the standpoint of supply chain diversification. At the third-quarter earnings call, Samsung said it has begun operating the cleanroom and moving manufacturing equipment into Taylor Fab 1, and that wafer starts are also under way to ramp up the production line. It plans to begin operating the fab in 2027 and strengthen global supply capacity.
The Taylor Fab 1 plan, however, has fallen substantially behind its original schedule. In the construction plan announced in November 2021, the investment was about $17 billion, with mass production targeted for the second half of 2024. The construction and equipment installation schedule was later changed; at the July 2026 earnings call the company expected operations to begin within that year, but the October 29 announcement says operations are scheduled to begin in 2027. This shows that building an advanced chip fab and ramping mass production can slip by years from the original plan.
For Taylor Fab 2, the July 2026 earnings call had already set out a policy of starting construction within the year and aiming for mass production in 2030. According to what Yonhap News reported on July 30, Samsung is preparing for construction of the second fab and plans to expand capacity according to demand. Korean outlet The Guru has also reported that the actual groundbreaking could come in November 2026. For the second fab, therefore, it is not that there is "no official plan"; rather, the construction policy the company has stated needs to be distinguished from the actual status of groundbreaking.
| Facility / investment plan | Samsung's disclosure | Progress and points to confirm |
|---|---|---|
| Taylor Fab 1 (US) | Operations to begin in 2027 | Cleanroom operation, equipment move-in and wafer starts for ramp-up under way. Delayed from the original target of mass production in the second half of 2024 |
| Taylor Fab 2 (US) | Targets construction start in 2026 and mass production in 2030 | Policy announced at July 2026 earnings call. Reports point to a November groundbreaking, but actual groundbreaking is unconfirmed |
| Additional fabs in Korea and the US | Investment under consideration given growing AI and HPC demand | Construction scale and groundbreaking schedule for individual new plans are not yet fixed |
| Advanced packaging facilities | Policy to strengthen logic manufacturing and packaging in an integrated manner | Developing technologies such as 3D stacking, but the specific capacity of additional facilities is undisclosed |
At the earnings call, Samsung also said it is considering investing in new fabs in Korea and the US to address the rapid growth of the AI and HPC markets. It intends to expand not only advanced logic manufacturing capacity but also advanced packaging capacity, building a production system that covers everything from wafer manufacturing to packaging. However, considering the construction of additional fabs is not the same as having decided specific investment amounts or groundbreaking dates. Plans to expand capacity at existing fabs and to build new fabs also need to be evaluated individually.
Even when a fab begins operating, it does not necessarily deliver the planned mass production capacity right away. In addition to installing lithography and etching equipment, process-to-process adjustment, optimization of manufacturing conditions and securing skilled engineers are required. For an advanced process such as 2nm in particular, there is a large gap between starting to run prototype wafers and being able to ship good chips stably at commercial scale. In evaluating the Taylor fab, it is important to check not only progress in construction and equipment move-in but also the actual start of mass production and the ramp in production capacity.
Operating Profit Above 100 Trillion Won, but Foundry-Only Profitability Remains Unconfirmed
To evaluate the recovery of Samsung's foundry business, consolidated results for the whole company need to be considered separately from the foundry's standalone profit and loss. According to the official third-quarter 2026 results, consolidated revenue reached 195.3 trillion won and operating profit 107.4 trillion won, both quarterly records. Quarterly operating profit exceeded 100 trillion won in a record performance, driven mainly by expanding demand for AI memory.
The Device Solutions (DS) division, which runs the semiconductor business, posted revenue of 147.1 trillion won and operating profit of 106.5 trillion won. Expanded sales of HBM4 and rising demand for server DRAM and SSDs in particular are lifting results. Samsung explained that the foundry business also improved profitability thanks to higher sales of HBM base dies and advanced processes, but the foundry's profit cannot be read directly from the DS division's overall operating profit.
Samsung does not disclose the foundry business's standalone operating profit, operating margin or absolute operating loss. The company's published figures alone therefore cannot show whether the business has already turned profitable or by how much it has narrowed its losses. Even if profitability "improved significantly" in the third quarter, that does not necessarily mean a return to an operating profit.
One input for estimating the size of the loss is brokerage forecasts for the non-memory business. In an analysis by Kiwoom Securities reported by Korean outlet Digital Today on September 29, 2026, the combined annual operating loss of the foundry and System LSI businesses was forecast to narrow from 6.742 trillion won in 2025 to 3.921 trillion won in 2026. The third-quarter 2026 operating loss was also seen shrinking to an estimated 777 billion won.
These, however, are analyst forecasts issued before the October 29 earnings release, not confirmed figures published by Samsung. They also cover the non-memory business combining foundry and System LSI, so they do not match the foundry's standalone profit and loss. Even if losses narrow as estimated, that is not the same as turning profitable. Advanced processes carry depreciation and R&D costs tied to enormous capital investment, so higher revenue does not immediately translate into sufficient profit.
The scale of capital expenditure is also important when considering the foundry business's room for growth. Samsung announced company-wide capital spending of about 84.4 trillion won for 2026, of which about 76.8 trillion won is planned for the DS division. In memory, it is investing in new fabs and other facilities to meet AI demand and medium- to long-term supply expansion, while in foundry it plans to increase investment centered on the Taylor fab. However, DS division investment also includes the memory business, and the specific allocation to the foundry business has not been disclosed.
The company-wide capex figure therefore cannot be tied directly to foundry capacity expansion. What will determine the business's profitability is not only whether fab capacity can be expanded in line with demand, but also whether the fixed costs that come with capital spending can be recovered through stable orders and mass production.
TSMC's 68% HPC Share Shows the Gap, and the Conditions for Judging Samsung's Recovery
The growth area Samsung emphasizes in its foundry business is the HPC market, including AI semiconductors. The company says it expects HPC to exceed 50% of foundry revenue in 2027. It aims to change its revenue mix by reducing its dependence on smartphone SoCs and expanding contract manufacturing of AI accelerators and data center chips.
TSMC, by contrast, has already secured large revenue in this area. According to its Q3 2026 earnings report, revenue from the HPC platform rose 22% quarter on quarter and accounted for 68% of net revenue. Smartphones accounted for 21%, making HPC the largest source of revenue. In addition, 2nm revenue reached 8% of total wafer revenue, and advanced processes including 3nm, 5nm and 7nm made up 79% of revenue.
What matters in comparing the two is that Samsung's HPC share of more than 50% is an outlook for 2027, whereas TSMC's 68% is based on revenue actually booked in the third quarter of 2026. Similarly, Samsung's metric of "2nm projects more than doubling year on year" and TSMC's "2nm at 8% of wafer revenue" measure different things. Samsung's order expansion is a sign of improving competitiveness in the advanced chip market, but there is not yet enough information to judge how far the gap with TSMC has narrowed in production scale or profitability.
TSMC's strength lies in combining advanced logic processes with advanced packaging technologies such as CoWoS to meet AI chip manufacturing demand. For AI accelerators, not only the logic chips that perform computation but also the packaging technology that connects them to HBM and integrates multiple chips is important. To be competitive in advanced contract chip manufacturing, therefore, scaling down transistor dimensions is not enough; packaging capacity and yield across the whole manufacturing process are also essential.
Samsung has strengths on this point as an integrated chipmaker that handles memory, logic and advanced packaging. In addition to making HBM base dies at its own foundry, it aims to expand contract manufacturing that combines multiple processes, using 2.5D packaging technologies such as I-Cube. It also plans to advance 3D stacking technology to strengthen its manufacturing base for AI chips. But owning these technologies is a separate matter from securing high yield and sufficient production capacity for external customers.
To judge whether Samsung's foundry recovery is real, several concrete developments need to be confirmed. First, it matters whether the second-generation 2nm mobile product, whose mass production starts in the fourth quarter of 2026, ships with stable quality and yield. Next comes whether mass production of 2nm products for the major external customer, planned for the second half of 2027, starts as scheduled and whether ordered projects lead to continuous wafer shipments. For the Faraday project too, what matters is not only the booking of design-related revenue but also the timing of the shift from prototyping and validation to mass production and the actual production scale.
The ramp-up of the Taylor fab is another factor that will shape the recovery. At Fab 1, equipment move-in and trial operation are advancing, but whether commercial production can get fully under way during 2027 depends on future progress. For Fab 2, it also needs to be confirmed whether the plan to break ground in 2026 and start mass production in 2030 is carried out. Future earnings calls will also be watched for whether losses in the non-memory business keep narrowing and how far the foundry's standalone profitability improves.
Samsung's forecast of utilization above 100% in 2027 is an important sign that demand for advanced contract chip manufacturing is expanding. The rise in 2nm projects and the Faraday order support the view that the company is making some progress in winning external customers. But order growth alone does not show that the foundry's rebuilding is complete. What will ultimately be tested is whether Samsung can secure stable yield on advanced processes and turn demand into continuous shipments and profit. The mass production of 2nm products and the ramp-up of the Taylor fab, both proceeding from late 2026 into 2027, will be the critical phase for judging success.