Samsung Foundry, which has been mired in yield problems and order losses for the past few years, saw a positive development at the recent NVIDIA GTC 2026 conference, spearheaded by NVIDIA CEO Jensen Huang.Public thanks to SamsungThis confirms that their new AI inference chip, Groq 3 LPU, is manufactured using their process technology. With the 2nm process yield stabilizing, coupled with orders from customers such as Tesla, AMD, and Qualcomm...Return or negotiationSamsung's foundry division has regarded the fourth quarter of this year as the decisive battle to "turn losses into profits".
NVIDIA embraces Samsung's manufacturing process again
In his keynote address at GTC 2026, Jensen Huang introduced the Language Processing Unit (LPU) from Groq, which he acquired at a high price last year, into the Vera Rubin rack system. He also mentioned that Samsung assisted in manufacturing this chip, which uses Samsung's 4nm (SF4X) process. This marks the second time that an NVIDIA product has been manufactured by Samsung since the GeForce RTX 30 series graphics cards, which also means that Samsung's process has passed NVIDIA's current stringent technical verification.
Market analysts believe that NVIDIA's core products, such as the Rubin GPU, are still manufactured using TSMC's advanced manufacturing technology. However, outsourcing the production of Grog chips to Samsung not only represents an improvement in Samsung's process yield but also helps to alleviate the pressure on TSMC's manufacturing capacity, thereby achieving a risk diversification effect. On the other hand, it also allows NVIDIA to maintain its cooperative relationship with Samsung.
2nm offensive launched: Tesla and AMD orders become the hope for profitability
In addition to its existing 4nm process orders, Samsung is further investing in its 2nm GAA (Gate All-Around) architecture process, and has successfully secured chip manufacturing orders from Tesla and AMD. Samsung's current foundry development status is as follows:
Stable yield:Industry experts point out that Samsung's 2nm process yield has reached a stable level, and the company has begun preparations for full-scale mass production of its flagship processor, the Exynos 2600.
Tesla AI chip:Samsung has reached a strategic partnership with Tesla to produce Tesla's next-generation self-driving chips (AI5/AI6) at its Taylor plant in Texas, USA, using a 2nm process.
AMD is in talks:Due to the extremely tight advanced production capacity of TSMC, AMD is currently in active talks with Samsung to cooperate on 2nm process manufacturing in order to mitigate the production risks of its chip products.
Taylor factory progress:Samsung’s Taylor plant in Texas is expected to begin production in the second half of 2026 and is expected to become an important local base for Samsung to win over major North American customers (such as Qualcomm and Microsoft).
Financial Outlook: The fourth quarter is expected to end consecutive years of losses, with a target profit of 2 trillion won.
Since 2023, Samsung's foundry business has been operating at a huge loss due to massive investments in advanced processes and customer attrition. However, with improved 4nm yields and the arrival of 2nm customers, Samsung has set a target of achieving break-even in the fourth quarter of 2026 and expects to turn a profit. Analysts predict that Samsung's foundry business will have the opportunity to challenge an operating profit of 2 trillion won (approximately NT$470 billion) for the full year of 2026.
