Samsung Miss – Why?
Samsung Electronics (005930.KS) reported preliminary 2Q results of 74t won ($53.77b US) in sales, 3.4% lower than consensus of 76.64t won ($55.685b US) and more importantly operating profit of 4.6t won ($3.34b US) against consensus of between 6.2t won and 6.35t won ($4.05b - $4.61b US). While the details of the miss will not be revealed until the call later this month, Samsung did give some information as to the poor results, which it acknowledged.
“Our preliminary results for 2Q25 fell short of market expectations. This explanatory note is provided to assist in the understanding of the key factors behind the decline and minimize confusion prior to the release of our full results in the 2Q25 earnings call on the 31st of July.
The DS Division recorded a quarter-on-quarter decline in profit due to inventory value adjustments and the impact of US restrictions on advanced AI chips for China. - The memory business saw a decline in earnings due to one-off costs such as inventory value adjustments, but improved HBM products are proceeding with customer evaluation and shipments for key customers. - The non-memory business experienced a decline in earnings due to sales restrictions and related inventory value adjustments stemming from US export restrictions on advanced AI chips for China, as well as continued low utilization rates. However, in the second half of the year, the operating loss is expected to narrow as utilization improves due to a gradual recovery in demand.”
Samsung’s High Bandwidth Memory production node for HBM4 differs from competitors SK Hynix (000660.KS) and Micron (MU) in that it uses a 1c process (6th Gen), as opposed to the 1b process (5th Gen) that Hynix and Micron use. It is certainly less mature, but is able to process line widths below 12nm, allowing for higher density features. That said Samsung has been unable to get the HBM3E1, its current memory product, a qualified at Nvidia (NVDA), a key customer and made a number of changes earlier this year to improve heat dissipation and reduce power consumption but was unable to qualify at Nvidia in May and is now working toward another evaluation in September. As Samsung has been in production for the HBM 3E, without Nvidia qualification, the company was forced to make a ‘value adjustment’ on the produced but not sold memory and the higher production costs that helped to create the 2Q miss.
Samsung expects to move ahead of its rivals with its HBM4 iteration (1c), for which it expects to begin sampling later this year for production in 2026. If successful, it will give Samsung the edge it needs to regain share and stature with major customers, especially Nvidia, but the poor results for the HBM3E does not lend much in the way of confidence. That said, Samsung installed a new head of the DS division in May of 2024 (officially appointed in March 2025), so his ability to right a listing ship will likely live or die on how well Samsung is able to finalize the HBM3E and transition to the HBM4 over the next12 to 18 months.
“Our preliminary results for 2Q25 fell short of market expectations. This explanatory note is provided to assist in the understanding of the key factors behind the decline and minimize confusion prior to the release of our full results in the 2Q25 earnings call on the 31st of July.
The DS Division recorded a quarter-on-quarter decline in profit due to inventory value adjustments and the impact of US restrictions on advanced AI chips for China. - The memory business saw a decline in earnings due to one-off costs such as inventory value adjustments, but improved HBM products are proceeding with customer evaluation and shipments for key customers. - The non-memory business experienced a decline in earnings due to sales restrictions and related inventory value adjustments stemming from US export restrictions on advanced AI chips for China, as well as continued low utilization rates. However, in the second half of the year, the operating loss is expected to narrow as utilization improves due to a gradual recovery in demand.”
Samsung’s High Bandwidth Memory production node for HBM4 differs from competitors SK Hynix (000660.KS) and Micron (MU) in that it uses a 1c process (6th Gen), as opposed to the 1b process (5th Gen) that Hynix and Micron use. It is certainly less mature, but is able to process line widths below 12nm, allowing for higher density features. That said Samsung has been unable to get the HBM3E1, its current memory product, a qualified at Nvidia (NVDA), a key customer and made a number of changes earlier this year to improve heat dissipation and reduce power consumption but was unable to qualify at Nvidia in May and is now working toward another evaluation in September. As Samsung has been in production for the HBM 3E, without Nvidia qualification, the company was forced to make a ‘value adjustment’ on the produced but not sold memory and the higher production costs that helped to create the 2Q miss.
Samsung expects to move ahead of its rivals with its HBM4 iteration (1c), for which it expects to begin sampling later this year for production in 2026. If successful, it will give Samsung the edge it needs to regain share and stature with major customers, especially Nvidia, but the poor results for the HBM3E does not lend much in the way of confidence. That said, Samsung installed a new head of the DS division in May of 2024 (officially appointed in March 2025), so his ability to right a listing ship will likely live or die on how well Samsung is able to finalize the HBM3E and transition to the HBM4 over the next12 to 18 months.
Figure 1 - 2025 YTD Quarterly Memory Market Producer Revenue Comparison (Bil. $ US) - Source: SCMR LLC, Counterpoint Memory Tracker
RSS Feed