2024.08.15News
Korean Media Analyzes Delay in HBM Supply Leading to "Contradictory Situation" - Is It a Win or Stain for Samsung?
Author: Yu-Jou Lin | Release Date: August 14, 2024 17:02 | Categories: AI Artificial Intelligence, Samsung, Memory Storage
Samsung has been struggling for over a year to develop High Bandwidth Memory (HBM). The latest market news has revealed that Samsung Electronics' 8-layer HBM3E chip has passed NVIDIA's tests. However, the entire situation is still quite mysterious as Samsung had previously clarified that the authentication process had not been completed before July.
At the same time, the latest AI chip from NVIDIA may face delayed shipments due to design flaws, causing concerns in the market for AI chips. Some reports suggested that even though Samsung is still in need of validation, NVIDIA may also need Samsung in return.
Based on the current market information, Samsung's HBM3E has already passed NVIDIA's validation, but both parties have not yet finalized the supply contract, with expectations of supply beginning in the fourth quarter. Reportedly, Samsung is currently unable to easily resolve the heating problem with HBM, but there is a high possibility that NVIDIA's validation will be successful because any further delays could pose risks for NVIDIA.
Employees of consulting firms believe that the price of each NVIDIA chip is high and the overall investment cycle in the industry is unlikely to exceed three years. NVIDIA also needs to consider its customers' budgets when planning its roadmap, which is also why it continues to give Samsung opportunities. It is rumored that Samsung is also evaluating the duration of the demand for AI server deployment internally.
The report pointed out that the deployment cost of AI servers at major companies such as Microsoft, Meta, and Google is about 40 times higher than that of general servers, with over 80% of it used for NVIDIA AI GPU. This investment is expected to exceed 250 trillion KRW just this year, but no one knows when it will be profitable, and macroeconomic factors could trigger a stock market crash. The report questioned whether consumers would still buy NVIDIA's next-generation semiconductor if large tech companies investing trillions of dollars in AI cannot get them to open their wallets.
The high prices of NVIDIA chips have prompted Samsung to enter the supply chain, which may help to adjust the prices.
The report suggests that this may be the reason for the delay in the release of NVIDIA's next-generation GPU. The industry believes that if NVIDIA needs to compromise and lower product prices, it must attract Samsung Electronics as an HBM3E supplier. Semiconductor industry insiders said, "NVIDIA's pace is too fast. They are trying to launch Blackwell two years after the Hopper series launch, other than SK Hynix with more experience, other companies are finding it difficult to flexibly handle sales strategies."
HBM orders are signed based on the supplier's capacity, and if there are deviations in the schedule, it is difficult to increase the capacity of other suppliers. In addition, SK Hynix has already announced full capacity for next year in January. If Samsung cannot join the supply lineup, it will limit the overall supply of AI chips in the market, potentially leading to increased prices.
The report stated that the outside world is worried about the overheating of AI market capacity and Samsung appears to be a counterbalance to NVIDIA's development roadmap and the entire AI industry. They even described Samsung's role in the AI market as a pacemaker. However, for Samsung, the memory powerhouse, this seems to be a stain.
Nevertheless, this is still an opportunity for Samsung to close the gap with its competitors. Starting with the 8-layer HBM3E, collaborating with NVIDIA will help stabilize the market and increase profitability while narrowing the gap with its rivals. However, if unsuccessful, the entire industry, including Samsung, will be affected. Industry insiders expect that next year's demand for HBM3E will account for over 80% of the total production.
Investors believe that whether it's competing with TSMC for wafer outsourcing or in the area of HBM, Samsung ranks second, and therefore, the external expectations for the company to cross the NVIDIA threshold are exceptionally high, and it must seize the opportunity.
Top