What lifted Samsung's foundry results was not its 2nm order book

Profitability at Samsung's foundry business improved sharply from a year ago, and the company said a turnaround is not far off. Utilization at advanced nodes of 8nm and below has reached peak levels, and better yields and price increases came on top of that.

Samsung Electronics Announces Second Quarter 2026 Results (Samsung Newsroom, 2026-07-30)

출처: news.samsung.com

The release published that same day says nothing about a foundry turnaround. It stops at the Device Solutions division, the chip arm, with revenue of 127.5 trillion won and operating profit of 89.2 trillion won. The foundry detail came out afterwards, on the earnings call.

So what actually lifted the revenue

The company named two reasons for higher foundry revenue. Demand for HBM base dies grew, and volume arrived from a North American customer's products.

Samsung Electronics to break ground on 2nd fab in Taylor within this year (The Korea Times, 2026-07-30)

출처: koreatimes.co.kr

A base die is the chip that sits at the bottom of an HBM stack. It is the layer that organizes signals between the stacked DRAM above it and the GPU outside, and it is built on a foundry line, not on a memory line.

So when memory sells well, foundry revenue rises with it. The part that improved because outside customers came back and the part that improved because Samsung's own memory sold well are mixed inside one number.

How far along is 2nm

Based on production experience with its first generation 2nm process, the company said it will begin the full ramp of second generation 2nm in the second half. In the second quarter it took 2nm work from one large cloud provider and from high performance computing customers, and it said talks with major customers including Broadcom are widening. It expects the number of 2nm projects won this year to more than double from last year.

When does output from today's construction arrive

The first Taylor fab is preparing to operate this year, and groundbreaking on the second is planned by year end with mass production targeted for 2030. With more inquiries about 1.4nm, the company is also reviewing securing one more fab.

Samsung to Break Ground on Texas Fab 2 by Year-End, Target 2030 Output (Seoul Economic Daily, 2026-07-30)

출처: en.sedaily.com

That article adds that the first Taylor fab began operating this year. So the gap from groundbreaking to output is four years, and that gap is the substance of the company's admission that advanced node expansion has not kept up with demand. It is the same structure as memory, where the company has said wafers from a new plant take more than three and a half years to appear. Capacity decided today does not solve today's shortage.

Where do the older processes go

Mature nodes are shifting away from volume competition toward products that can command a higher price. The company said it is adding capacity for 8nm and 17nm, image sensors, and the nonvolatile memory embedded inside chips. It also named silicon capacitors and silicon photonics as areas it is growing.

So what will settle it

The question comes down to one thing. Is the foundry recovery outside customers returning, or Samsung's own memory pulling it along. If 2nm project wins really do more than double this year and Broadcom volume shows up as a contract, the first reading holds. If the numbers keep improving while the customer list does not lengthen, the foundry cools together with memory once the memory boom cools. Running memory and foundry inside one company was called a weakness for years. This quarter that structure held the foundry up. The same structure works in the other direction too.