[Chip Talk] Samsung Foundry Sheds 'Loss-Making' Label with Spectacular Comeback... Big Tech Orders Pour In

Following Tesla and Qualcomm, Now Nvidia and AMD
Consecutive Orders Put Q4 Profitability Within Reach
"Groq3" to Be Contract-Manufactured at Pyeongtaek Campus in Q3
Talks Underway for Contract Manufacturing of AMD’s Next-Generation Chips

Samsung Electronics’ foundry (semiconductor contract manufacturing) division is making a spectacular comeback this year, riding the semiconductor supercycle. After securing large-scale orders from Tesla and Qualcomm, it is now producing Nvidia’s next-generation artificial intelligence (AI) chip, the “Groq 3.” With discussions now underway to contract manufacture AMD’s next-generation chips as well, there is growing anticipation that the foundry division—long considered the “problem child” of the Device Solutions (DS) segment—will finally shed its reputation for being “perpetually in the red” and swing to profit in the fourth quarter of this year.

According to the semiconductor industry on March 21, Samsung Electronics’ foundry division has established its readiness for a full-fledged turnaround to profitability, thanks to its technological and infrastructure competitiveness. The yield of its 2-nanometer process has stabilized, and with the U.S. Taylor plant scheduled to begin operations in the second half of the year, production capacity is expected to improve significantly.

Internally, the target for the foundry division’s return to profitability has reportedly been set for the fourth quarter of this year. Securities analysts also predict that Samsung’s foundry division will substantially reduce its quarterly losses to several hundred billion won starting in the second half of the year, and may achieve profitability as early as the fourth quarter.

The foundry business had previously suffered from low profitability, due to the need for substantial upfront investment in advanced processes. The adoption of extreme ultraviolet (EUV) lithography equipment and the burden of investment in cutting-edge processes resulted in quarterly losses of around 1 trillion won since 2022.

However, the situation has changed recently. Yields in advanced processes—ultra-fine manufacturing processes for semiconductor circuits, once considered a chronic issue—have stabilized, rapidly increasing utilization rates. The operating rate of the Pyeongtaek foundry line is said to have risen from around 50% last year to over 90% recently.

Most notably, since the second half of last year, Samsung has secured major contract manufacturing deals from large clients, firmly laying the foundation for a rebound. Jensen Huang, CEO of Nvidia, announced on March 16 (local time) that Samsung’s foundry division will produce Nvidia’s inference-dedicated chip, the “Groq3 Language Processing Unit (LPU).”

The Groq3 LPU is a chip that splits roles with Nvidia’s “Rubin” graphics processing unit (GPU) to enhance inference performance and efficiency. CEO Huang explained that this chip will be installed in Nvidia’s next-generation AI chip, the Vera Rubin system, adding, “Shipments will likely begin in the second half of this year, possibly around the third quarter.”

Nvidia, a fabless (semiconductor design) company, has historically entrusted the majority of its chip production to TSMC, the world’s largest foundry company. However, with the announcement that greater manufacturing scale is needed, Samsung Electronics has now been officially cited as a key partner not only for high-bandwidth memory (HBM) but also in the foundry sector. Samsung Electronics is producing the Groq3 using its 4-nanometer (nm; one billionth of a meter) process at the Pyeongtaek campus, and is expected to begin shipments as early as the second half of this year.

In July last year, Samsung secured a next-generation AI chip order from Tesla worth as much as 24 trillion won. Tesla CEO Elon Musk revealed that Samsung’s Taylor plant in Texas, USA, will manufacture Tesla’s next-generation AI6 chips. Discussions are also underway with Qualcomm for contract manufacturing using the 2-nanometer process. Han Jinman, President and Head of the Foundry Business at Samsung Electronics, stated, “We have received more orders than expected,” adding, “If the effectiveness of the Groq LPU is confirmed, we expect demand to increase significantly.”

On top of this, discussions on advanced process collaboration with AMD are also materializing. On March 18, Samsung Electronics signed a memorandum of understanding (MOU) to expand cooperation in next-generation AI memory and computing technologies during a meeting with AMD CEO Lisa Su. The agreement not only confirmed the supply of 6th-generation high-bandwidth memory (HBM4), but also expanded cooperation to the foundry sector. Currently, most AMD chips are produced by TSMC, but there is now a significant possibility that some will be manufactured by Samsung Foundry.

An industry insider commented, “Samsung Electronics differentiates itself from competitors like TSMC by providing a one-stop solution covering HBM, foundry, and packaging. Despite continued losses in the past, the technical capabilities gained through ongoing investment are now translating into the acquisition of major clients.”

This content was produced with the assistance of AI translation services.