[SEOUL] Samsung Electronics’ profit shrank more than expected after encountering hiccups in plans to sell cutting-edge artificial intelligence (AI) memory to Nvidia.
South Korea’s largest company reported preliminary operating profit of 4.6 trillion won (S$4.3 billion) in the June quarter, a roughly 56 per cent drop from a year ago and its first profit decline in more than a year. Analysts on average had projected a 41 per cent decline. Inventory-related costs contributed to the drop, Samsung said.
Revenue stood at 74 trillion won. Samsung will provide a full financial statement with net income and divisional breakdowns later this month.
Samsung has been struggling to regain its footing in high-bandwidth memory (HBM) chips, which are critical for powering Nvidia’s AI accelerators. The company has yet to secure certification from Nvidia for its most advanced product, the 12-layer HBM3E, creating an unusually long lead time for rival SK Hynix in the highly lucrative space. Meanwhile, US competitor Micron Technology has been rapidly advancing to stake its own claim.
Analysts polled by Bloomberg News prior to the preliminary earnings release expected Samsung’s chip division to post an operating profit of 2.7 trillion won in the second quarter, up from 1.1 trillion in the prior quarter but still significantly lower than 6.5 trillion won a year earlier.
In April, Samsung signalled a better outlook, saying it shipped enhanced HBM3E samples to major customers and expected that product line to contribute to revenue in the second quarter. The company also said it plans to begin mass production of HBM4 chips in the second half of the year.
BT in your inbox
Start and end each day with the latest news stories and analyses delivered straight to your inbox.
Samsung is fighting to catch up to SK Hynix, which has aggressively positioned itself as Nvidia’s primary HBM4 supplier. It shipped the world’s first 12-layer HBM4 samples to customers ahead of schedule, followed by Micron in June, while Samsung has had to revise its 12-layer HBM3E design.
Samsung secured an order from Advanced Micro Devices, joining Micron as a supplier, according to a June release. But its failure to win early certification for HBM3E chips from Nvidia – the dominant maker of AI-supporting graphics processing units – is hurting its attempts to take significant market share.
Bernstein analysts led by Mark Li, who had previously expected Samsung’s 12-layer HBM3E would be qualified by Nvidia in the second quarter, trimmed their forecast for Samsung’s HBM market share, saying they now expect certification in the third quarter.
“Samsung will gradually narrow the gap vs rivals,” they wrote in a Jun 23 research note. “We forecast SK Hynix remains the leader in 2027, but with others catching up and SK Hynix’s edge eroding, the shares held by suppliers will be more similarly distributed than now.”
Bernstein estimates SK Hynix holds 57 per cent of the HBM market in 2025, followed by Samsung at 27 per cent and Micron at 16 per cent.
At its annual shareholder meeting in March, Samsung vowed to strengthen its position in the HBM market this year, responding to concerns over its underperformance in AI. Jun Young-hyun, head of Samsung’s chip business, said that Samsung’s failure to secure an early lead in the HBM market contributed to it lagging behind rival SK Hynix and pledged not to repeat the mistake with HBM4. The next-generation memory is expected to be used in Nvidia’s Rubin GPU architecture.
“While uncertainties remain around the timeline for quality certification, we see no indication of a strategy shift and believe the company is on track for a 3Q2025 market launch,” Daishin Securities analyst Ryu Hyung-keun wrote in a recent research report. BLOOMBERG
