Samsung Electronics is riding the AI wave with an increase in sales and a huge jump in operating profit thanks to a recovery in the memory market, and the firm expects demand from server AI to stay strong for the rest of the year.

Meanwhile, Samsung's system LSI business chief, Tommy Kwon, confirmed that the next iteration of the company's flagship Exynos chip, which uses a 3nm fabrication process, will launch next year. Kwon also confirmed the nomenclature, until now speculative, on the analysts' call last night: "For our business, our industry's first 3-nanometer SoC for wearables is receiving strong, positive initial feedback. With this, we plan to focus our resources on ensuring a stable supply of the Exynos 2500 for flagships products."

Kwon didn't say which, but the chip is expected to power 2025's Galaxy S25 and Galaxy S25+, depending on your location. Korea's news outlets as well as IT tipsters have claimed that Samsung believes it will outperform Qualcomm's rival, but the only feeds and speeds currently available are leaks and rumors.

The Korean semiconductor giant reported [PDF] an increase in total sales of 23 percent year-on-year to ₩74.07 trillion ($53.8 billion) for its Q2 2024 that ended June 30, helped by memory sales that were up 142 percent to ₩21.74 trillion ($15.8 billion).

As predicted, the company's operating profit saw a fifteenfold increase, reaching ₩10.44 trillion ($7.6 billion) from just ₩0.67 trillion ($488 million) for the same period last year, closely matching the guidance Samsung issued earlier this month.

Samsung said its gross profit this quarter was ₩29.8 trillion ($21 billion), up ₩3.7 trillion ($2.7 billion) compared to last quarter, mainly due to greater demand for DDR5, high bandwidth memory (HBM) and other high value-added memory products.

"The memory market in the second quarter continued to recover due to strong generative AI server demands," said Jaejune Kim, EVP for Samsung's Memory business, speaking on a conference call to discuss the results.

"Our robust demand for not only HBM but also conventional DRAM and SSD for server drove the market recovery, with contributions also coming from the expansion of AI investments by cloud service providers and rising demand for AI in the enterprise on-premise market," Kim added.

HBM sales increased by more than 50 percent, while server DDR5 sales saw a greater than 80 percent rise, according to Kim, thanks to increased shipments and prices rising because of high demand. Server SSDs also saw a revenue increase of more than 40 percent, he added.

"In addition, we strengthened our leadership in the DDR5 market by starting mass volume business of 128 GB product with customers based on 1b nanometer 32 Gb DDR5, which was developed for the first time in the industry," Kim claimed.

The company expects AI-driven memory demand to continue during the second half of 2024, and hinted it may increase production to match.

"We foresee a solid server demand led by generative AI to lead the strengths. Therefore, we also plan to actively respond to the demand for high value-added products for AI in the second half," Kim said.

But it wasn't all about memory, and Samsung said its display revenue increased by 42 percent off the back of robust OLED sales. However, the company's MX division (mobile business) saw a fall in revenue of 19 percent, which it blamed on weak seasonal demand for smartphones.

Samsung's Foundry business also claimed double-digit growth compared to the first quarter, and the company said the volume of AI and HPC customers had doubled year-on-year due to expanded orders in sub-5nm production lines.

Foundry VP Taejoong Song said that the process design kit for Samsung's upcoming 2nm GAA (gate all around) production node had its "upgrade and diligence to several customers" to allow them to move forward with designs for mass production in 2025.

In response to a question, Samsung said it had experienced no issues in responding to customer demand from the strike action by workers that started in June.

Samsung also addressed questions that kept surfacing recently regarding its HBM memory for Nvidia GPUs, but didn't really settle the issue.

"While I understand that there is significant interest among investors and media regarding our HBM qualification status, we are under NDA commitment with our customers, and so we seek understanding that we're not able to comment on those details," Kim said.

However, Kim did say that Samsung is expanding production in order to supply all major GPU customers, and that the company is on track for a market launch of HBM4 memory in the second half of 2025. ®

https://www.theregister.com//2024/07/31/samsung_q2_2024/