Samsung Electronics has taken a new step forward in its collaboration with NVIDIA. Beyond supplying high-bandwidth memory (HBM) for NVIDIA’s artificial intelligence (AI) accelerators, the company is actively exploring opportunities to secure orders for next-generation chips via its advanced foundry processes. The cooperation covers NVIDIA’s Drive AGX Thor chips for autonomous vehicles and Groq Language Processing Units (LPUs).

     On June 8, Jun Young-hyun, Vice Chairman of Samsung Electronics, attended the NVIDIA-hosted Korea AI Ecosystem Reception and held a private business meeting with NVIDIA CEO Jensen Huang that evening. After the meeting, Jun noted that this was one of the most in-depth talks between the two partners since their cooperation began. The executives discussed the supply and foundry collaboration for the upcoming sixth-generation HBM4 in the short term, as well as medium and long-term joint development initiatives.

     Samsung Electronics stressed that it will ensure sufficient supply of HBM4 and low-power memory module SOCAMM, both set for mass production this year. The company also plans to sustain long-term cooperation starting next year, covering the seventh-generation HBM4E and eighth-generation HBM5. Former Vice Chairman Jun Ji-hyun revealed that Samsung is forging deeper foundry ties with NVIDIA. The two sides are exploring the production of NVIDIA’s autonomous driving chips using Samsung’s 4nm and 8nm processes. Additionally, Samsung Foundry is partnering with NVIDIA on another accelerator, Groq, while discussing collaboration on next-generation accelerators.

     NVIDIA has outsourced the production of its third-generation Groq LPU (LP30) to Samsung Foundry, which manufactures the chip on a 4nm process. The LP30 is a product launched by NVIDIA after its USD 20 billion (approximately 30.6 trillion South Korean won) acquisition of Groq last December. Jun Jeong-hoon, Vice Chairman of Samsung Electronics, confirmed that Samsung Foundry will also produce the next-gen LP40, dismissing earlier rumors that TSMC would land the LP40 order. He added that Samsung Electronics will fully support NVIDIA’s growth and serve as its reliable partner.

     Notably, Samsung Electronics’ foundry division is on track for a performance turnaround. Internal sources indicate the unit could return to profitability as early as the third quarter of this year. Analysts attribute the improved profitability mainly to rising output of base chips built on the advanced 2nm process and increased production of high-bandwidth memory (HBM).

     Driven by improved yield rates and large-volume orders, Samsung’s foundry business is expected to swing into the black ahead of schedule in Q3 2026. This will put an end to trillions of South Korean won in annual losses that have persisted since 2022, marking a business recovery after roughly four years. The performance of Samsung’s advanced manufacturing processes has exceeded expectations. As of Q1 2026, the yield rate of its 2nm Gate-All-Around (GAA) process has surpassed 60%. While still below the 70% threshold for cost-effective mass production, the yield is adequate for initial volume production and helps attract new clients. Financial reports released last April showed robust utilization rates at Samsung’s advanced process production lines, delivering double-digit year-on-year revenue growth even in off-seasons.

     Customer expansion efforts rolled out since last year are set to translate into full revenue growth in the second half of 2026. In July last year, Samsung signed a long-term supply contract worth 22.8 trillion South Korean won with Tesla for autonomous driving chips. In the latter half of this year, its plant in Taylor, Texas, will begin mass production of 2nm-process AI5 and AI6 chips. During NVIDIA’s GTC 2026 conference last March, Jensen Huang publicly confirmed the foundry partnership with Samsung Electronics while unveiling the Groq-3 chip.

     Order volume is expanding rapidly. Samsung forecasts that orders related to its 2nm process will surge by over 130% year-on-year in 2026. The company is also expanding cooperation with major tech players including Apple and Nintendo. All allocation for the sixth-generation HBM (HBM4) has been fully sold out.

     The upcoming mass production at Samsung’s Taylor wafer fab will help optimize its cost structure. The USD 37 billion (approximately 54 trillion South Korean won) project previously exacerbated losses for the foundry division due to massive construction costs. Once the fab goes into full operation in the second half of the year, depreciation expenses will dilute fixed costs and quickly improve profit margins.

     Furthermore, skyrocketing demand for AI accelerators has pushed TSMC’s advanced process production lines to near full capacity, creating favorable conditions for Samsung. Many fabless chip companies are considering Samsung as an alternative supplier. Meanwhile, AMD is mulling a dual-foundry strategy for some of its next-generation graphics processing units (GPUs).

This article is reposted from DZSC Electronics Market Network. Source: https://www.dzsc.com