The latest performance forecast released by Samsung Electronics, the South Korean technology giant and the world's largest memory chip manufacturer, shows that the profit scale in the third quarter is lower than the market consensus. In a performance preview released on Tuesday, the world's top memory chip maker said operating profit for the latest quarter ending in September is expected to be about 9.10 trillion won, a 274% surge from last year's 2.43 trillion won.
However, the figure fell short of consensus estimates among economists compiled by LSEG. Against the background of the recent surge in South Korea's chip exports and the sharp decline in chip inventories due to the surge in storage demand brought about by the global AI boom, Samsung's operating profit expectations have been very disappointing to analysts and chip stock investors.
It is understood that analysts covered by LSEG generally expect Samsung Electronics’ quarterly operating profit as of September 30 to be approximately 11.456 trillion won (approximately US$7.7 billion). Since Samsung’s performance for many quarters has shown that HBM and data center DRAM and NAND failed to achieve huge profit increases like competitors SK Hynix and Micron, some analysts suspect that Samsung has failed to catch up with this unprecedented AI boom. According to analyst expectations compiled by LSEG, Samsung's total revenue in this quarter is expected to reach approximately 81.96 trillion won (approximately US$61 billion).
Jun Young-hyun, vice chairman of Samsung and the new head of the device solutions department, even issued a rare apology statement after the release of the performance forecast, and said that he was working hard to solve major problems related to artificial intelligence chip products such as HBM. Samsung said in a statement: "Performance of Samsung's memory chip business, which includes HBM and enterprise SSDs, declined due to 'one-time costs and significant negative impacts,' including inventory adjustments by mobile customers and increased supply of traditional storage products from other memory chip companies in Asia."
South Korea is home to two of the world's largest memory chip manufacturers - SK Hynix and Samsung. Samsung is the world's leading manufacturer of memory chips and the world's largest manufacturer of memory chips used in equipment such as laptops and servers. Samsung Electronics is also the world's second-largest player in the smartphone market. In addition, Samsung has recently been striving to become one of the latest generation HBM3E suppliers of NVIDIA's most demanding Hopper architecture and the latest Blackwell architecture AIGPU.
The AI craze that has swept global enterprises since 2023 has led to a surge in demand for AI servers. The world's top data center server manufacturers such as Dell and Super Micro usually use Samsung and Micron data center DDR series products in their enterprise-level AI servers, and Samsung/Micron SSD, one of the mainstream applications of NAND storage, is widely used in the server main storage system of computing systems, while SK Hynix HBM storage systems are fully integrated with NVIDIA AIGPU. This is also an important logic behind the surge in demand for HBM storage systems and the entire DRAM and NAND storage.
Judging from South Korea's chip export data and chip inventory data, the explosive demand for memory chips becomes clearer. Data released by the South Korean government show that despite the slowdown in growth, semiconductor exports in September still increased significantly by 37% year-on-year, increasing for 11 consecutive months, which was slightly weaker than the 38.8% increase in August.
In August, South Korea's chip product inventories shrank at the fastest pace since 2009, indicating continued demand for high-performance memory chips used in artificial intelligence development. Data show that South Korea's chip inventory fell by 42.6% from the same period last year, which was higher than the 34.3% decline reported in July. Production and shipments increased 10.3% and 16.1% respectively, further indicating that the global chip industry boom cycle continued for most of the third quarter.
As the scale of new construction and expansion of global data centers has continued to expand in the past year due to the surge in demand for AI computing power brought by applications such as ChatGPT and MetaAI, the enterprise-level DRAM and NAND fields have become the core driving force for the performance growth of global memory chip companies. Samsung is indeed an important leader, but its position does not have the absolute dominance it does in consumer electronics fields such as smartphones and PCs. Compared with the consumer market, the competitive landscape of the enterprise market is more complex, with major players including Micron, SK Hynix, Western Digital, and NetApp. Micron and SK Hynix perform more prominently in the enterprise SSD market than Samsung, especially in the field of customized data center storage solutions.
Regarding the HBM storage system that is crucial for artificial intelligence large model training/inference and NVIDIA H100, H200 and the latest Blackwell architecture AIGPU hardware system, the company added that there have also been unexpected delays in shipping high-bandwidth memory "HBM3E" storage systems to major customers (most likely NVIDIA and AMD), further validating market speculation that the largest memory chip manufacturer currently has no HBM flagship product that can be produced, and Samsung may not be able to ride on this surge in HBM demand.
The HBM storage system is used in conjunction with the H100/H200/GB200AIGPU, the core hardware provided by AI chip leader Nvidia. HBM and AIGPU are indispensable for driving major artificial intelligence applications such as ChatGPT and Sora. Due to the almost never-ending market demand for NVIDIA's full line of AIGPU products, NVIDIA has become the world's most valuable chip company. The HBM storage system can provide information faster and help computing systems develop and run large artificial intelligence models.
NVIDIA H200AIGPU and the latest HBM based on Blackwell architecture B200/GB200AIGPU are equipped with the latest generation HBM storage system produced by SK Hynix - HBM3E. Another major HBM3E supplier is the storage giant Micron from the United States. Micron HBM3E will most likely be equipped with NVIDIA H200 and the latest B200/GB200AIGPU with extremely powerful performance.
HBM is a high-bandwidth, low-energy storage technology specifically used in the fields of high-performance computing and graphics processing. HBM uses 3D stacked storage technology to fully connect multiple stacked DRAM chips together and transmit data through micro-Through-SiliconVias (TSVs) to achieve high-speed and high-bandwidth data transmission. HBM uses 3D stacking technology to stack multiple memory chips together, which not only greatly reduces the space occupied by the storage system, but also greatly reduces the energy consumption of data transmission. High bandwidth can significantly improve data transmission efficiency, allowing large AI models to run more efficiently 24 hours a day.
In particular, the HBM storage system also has powerful low-latency features and can quickly respond to data access requests. Generative AI large models such as GPT-4 usually require frequent access to large data sets and extremely heavy large model inference workloads. The powerful low-latency feature can greatly improve the overall efficiency and response speed of the AI system. In the field of AI infrastructure, HBM storage systems are fully bound to NVIDIA H100/H200 AIGPU server systems, as well as NVIDIA B200 and GB200 AIGPU server systems that are about to be delivered in batches.
"Well, let's look at the numbers - it's very disappointing indeed," said Daniel Yoo, head of global asset allocation at Yuanta Securities in South Korea. He also noted that demand for traditional memory chips used in PCs and smartphones, which Samsung relies on, has not increased significantly globally.
"Samsung is not as aggressive as we have seen in the past in grabbing memory chip market share in the emerging area with the strongest storage demand - AI infrastructure. I think that is the big problem we are seeing," Yoo added.
"The company needs to maintain flexibility in its storage product supply system control, as the decline of the traditional DRAM market focused on personal products may cause Samsung more harm than its smaller competitors." Analysts from Macquarie Securities Research said in a recent report. DRAM is usually used in smartphones and personal computers, while enterprise-level traditional DRAM is used in fields such as enterprise servers.
Some media reported in September, citing two people familiar with the matter, that Samsung had instructed its global subsidiaries to reduce employees in certain departments by about 30%. Samsung's move was aimed at providing ample cash flow support for the successful launch of HBM with higher performance and in line with Nvidia's flagship AIGPU supply qualification, as well as enterprise-class SSDs with stronger demand than consumer electronics. Some media even reported that in order to speed up core technical work such as HBM R&D, manufacturing and all-round testing, Samsung executives even forced storage line employees to work overtime.
LSEG statistics show that Samsung Electronics' stock price listed on the Korean Stock Exchange has fallen 22% so far this year. The company will release detailed third-quarter results later this month. Samsung Electronics shares fell 0.98% after issuing guidance.