Key Takeaways
1. Samsung Foundry is struggling with low production yields and complications in HBM production, impacting its advanced process nodes.
2. Samsung received approval to supply Nvidia with HBM3E memory for AI chips, but it still trails behind SK Hynix, the main supplier for Nvidia.
3. Samsung missed an opportunity to collaborate with Nvidia in 2018, which could have positioned them as a key partner in HBM production and AI chip development.
4. Nvidia’s reliance on SK Hynix for HBM has contributed to a significant increase in SK Hynix’s stock price, while Samsung has lost out on AI chip manufacturing opportunities.
5. Samsung is now working to secure HBM business from Nvidia and AMD, aiming to provide HBM4 by early 2026.
It’s no secret that Samsung Foundry is having a tough time. The company is said to be facing challenges with low production yields in its advanced process nodes, which has also affected the manufacturing of its upcoming 1.4 nm node. Additionally, there are complications surrounding High-Bandwidth Memory (HBM) production.
Partnership Dynamics
Recently, Samsung has received approval to provide Nvidia with HBM3E memory for AI chips after missing out on the approval last year. Despite this, the company still lags behind SK Hynix, which is the primary supplier of HBM for Nvidia. The situation could have been quite different if Samsung had not turned down Nvidia’s offers back in 2018.
A new report reveals that Nvidia’s CEO, Jensen Huang, sought to expand the HBM partnership with Samsung in 2018. Huang reportedly visited Samsung to discuss collaboration in three main areas: HBM production, the development of new process nodes, and cooperation on CUDA. Such a partnership could have positioned Samsung as a pivotal ally for Nvidia.
Missed Opportunities
However, Samsung declined Nvidia’s proposals. Huang expressed his frustration, saying, “there was no one at Samsung to discuss long-term strategy with me.” This inability to engage in strategic discussions may have been linked to an ongoing criminal investigation involving Samsung Electronics Chairman Lee Jae-Yong at that time.
Nvidia has significantly benefitted from the AI boom due to its AI chips, which have seen remarkable demand. HBM is critical for these AI superchips, as components like the H200 require substantial high-throughput memory for AI inference tasks.
Market Impact
A large portion of Nvidia’s HBM chips comes from SK Hynix, which has greatly benefitted from this arrangement. At the present moment, SK Hynix’s stock price has surged more than 220% over the past five years.
Samsung’s choice to not collaborate with Nvidia may have repercussions for Samsung Foundry as well. Currently, TSMC manufactures all of Nvidia’s AI chips, including the latest B200 GPU. There have been speculations that Nvidia could engage Samsung to fabricate its upcoming gaming chips using the 2 nm process, but there is no concrete news about Samsung making Nvidia’s AI accelerators.
With Intel facing its own troubles and Samsung struggling to keep pace with TSMC, TSMC has gained a monopoly on Nvidia’s AI GPUs.
Future Prospects
While it may be too late for Samsung to fully counteract the consequences of its past decisions, the company is now working to secure some HBM business from Nvidia and AMD. Reports suggest that Samsung is aiming to provide HBM4 to both companies in early 2026.
Source:
Link
Leave a Reply