Over the past couple of years, Samsung has lost two big-name chip clients—Nvidia and Qualcomm—to TSMC due to power consumption issues. However, Nvidia could come back to Samsung Foundry in the future, as it doesn't want to rely too much on TSMC for its future chips. This is great news for Samsung.
Nvidia's future GPUs could be made by Samsung Foundry
According to a report from The Information, Nvidia is exploring a partnership with Samsung to manufacture new gaming chips (GPUs). These chips are simpler to make compared to Nvidia's AI accelerator chips, which are in huge demand and made by TSMC. The report claims that Nvidia is currently negotiating with Samsung to get discounted pricing. The world's biggest AI chip firm is reportedly aiming to get a 20-30% discount compared to TSMC's pricing.
The report claims cracks are appearing in Nvidia's relationship with TSMC, as the former's Blackwell chips faced some issues. Nvidia's spokesperson said it made changes to Blackwell chips to “improve production yield and aligned our schedule accordingly.” The chips that will be manufactured for the next four quarters have already been purchased by big tech firms, including AWS, CoreWeave, Google, Meta, Microsoft, and Oracle.
Another big consideration is the pricing for manufacturing these advanced chips. TSMC has increased its pricing over the past few years for advanced process nodes, and customers have no option but to accept higher pricing since TSMC is the world's most advanced chip maker. However, Nvidia is reportedly considering Samsung Foundry, which is the second-biggest chipmaker.
Samsung had multiple clients for its 5nm, 7nm, and 8nm process technologies. However, it didn't get enough clients for its 3nm and 4nm technologies, and the company is hoping to get clients back in the near future. While Samsung Foundry was the first in the world to make 3nm chips, it didn't get any big-name clients due to yield-related issues. Hopefully, the South Korean firm manages to improve its yield to serve customers like Qualcomm and Nvidia.
