SK Hynix logs record profit in Q2, sees stronger H2 ahead

SK Hynix Inc., the largest supplier of artificial intelligence chips to Nvidia Corp., reported its highest-ever quarterly results, with double-digit gains in both operating profit and revenue.

The earnings are twice those of Samsung Electronics Co. for the same quarter, in contrast to the bigger rival, the operating profit of which was halved.

A push by global Big Tech firms to launch new AI models with enhanced inference capabilities in the second half has fueled demand for 12-layer HBM3E, the most advanced AI memory chip currently in mass production.

The South Korean chipmaker offered a stronger outlook for the second half as its clients are slated to release new high-performance, high-capacity AI models, along with growing competition from countries in their own quests to build sovereign AI infrastructure.

Its operating profit soared 68.5% to 9.21 trillion won ($6.7 billion) in the second quarter from the year prior on a consolidated basis, according to its regulatory filing on Thursday.

Compared to the first quarter, the earnings jumped 23.8% and marked its highest-ever quarterly results since the 8.1 trillion won recorded in the fourth quarter of 2024. Its operating profit margin was 41%.

Revenue at the world’s top producer of high-bandwidth memory (HBM), key components for AI applications, spiked 35.4% on-year to 22.23 trillion won. Quarter on quarter, it climbed 26.0%.

The above-consensus preliminary earnings are double the 4.6 trillion won reported by Samsung Electronics for the April-June quarter, which represented a 55.9% plunge on-year and 31.2% drop from the prior quarter.

SK Hynix aims to double both the performance and production volume of HBM3E this year, compared to 2024. The company also said during its second-quarter earnings conference that it will deliver the more advanced model HBM4 on schedule.

Early this year, SK Hynix became the world’s first chipmaker to deliver an HBM4 sample to Nvidia, the world’s largest AI chip maker. Industry observers expect SK Hynix to begin mass production of HBM4 in the second half of this year.

The Korean semiconductor company is also leading the customized HBM market. It has already secured Nvidia, Microsoft Corp. and Broadcom Inc. as key clients of HBM4E, according to industry sources last month.

HBM4E, the seventh generation of HBM, is expected to be SK Hynix's first AI memory chip customized for individual clients.

SERVER MEMORY

To strengthen its position in the server AI chip market, SK Hynix plans to begin supplying LPDDR-based modules, or low-power memory for AI servers in data centers, later this year. LPDDR stands for low-power double data rate.

It is also preparing a higher-capacity 24 gigabit (Gb) GDDR7 for use in AI graphics processing units, up from the current 16Gb version.

Its net profit spiked 69.8% on-year in the first quarter, but fell 13.7% from the previous three months.

Write to Yeonhee Kim at yhkim@hankyung.com

Jennifer Nicholson-Breen edited this article.