Samsung posts record profit as AI memory demand reshapes chip market

Samsung Electronics divulged on Thursday (July 30) that it earned KRW 89.5 trillion in operating profit in the second quarter, boosted mainly by memory chips used in AI data centers. Its consolidated revenue increased significantly to KRW 171.5 trillion, an increase of 28% on the previous quarter, per the company’s earnings report.

The timing is significant as SK hynix and Micron Technology have both come up with reports of record quarters within days of each other, reiterating the same theme that the boom in AI infrastructure is increasingly benefiting the companies providing the memory used in sophisticated AI systems.

The findings also bring to the fore another issue for investors. There is no doubt that artificial intelligence has changed the demand for memory but the question is now whether price, supply limits and consumer expenditure will continue to be sufficient to support the semiconductor valuations.

Memory is carrying Samsung, mobile is not

Samsung’s financial performance proves how much AI has impacted its business model.

The Device Solutions division, which includes Samsung’s memory business made KRW 127.5 trillion in revenue and KRW 89.2 trillion in operating profit, almost all of the company’s earnings. Samsung pointed to the strong demand for high-end products such as high-bandwidth memory (HBM), server DRAM and enterprise SSDs used by AI cloud providers.

The outcome was not as encouraging for the mobile phone division. The Mobile eXperience and Networks unit generated KRW 33.2 trillion in sales while suffering an operating loss of KRW 0.7 trillion due to losses associated with increasing costs of materials. There were, however, gains for shareholders as the earnings per share rose to KRW 10,849, recording a 52% increase from previous years thanks to good earnings from semiconductors that compensated for the poor performance of the mobile phone business.

On July 27, TrendForce published an earnings preview indicating that Samsung’s earnings rose by approximately 18 times from last year during the same quarter, indicating that AI memory has rapidly changed the company’s performance dramatically.

SK hynix and Micron confirm the pattern

The performance of Samsung comes within a larger pattern in the industry.

The largest HBM firm, SK hynix, obtained KRW 79.3 trillion in revenue for the quarter and KRW 60.5 trillion in operating profit, obtaining an operating margin of around 76%. Revenue increased by 257% from year to year, and operating profit increased by 557%. In addition, the company has surpassed KRW 100 trillion in revenue for the first half for the very first time.

Toward the end of June, Micron wrapped up its report with impressive revenues of $41.46 billion for the third fiscal quarter and GAAP net income of $28.24 billion. The US-based memory manufacturer also pointed out the expansion of its multi-year supply contracts with hyperscale customers that grants them greater visibility into future AI-driven demand.

The strong execution by our teams enabled us to deliver record revenue and EPS. Memory is becoming increasingly strategic in the AI era.

Sanjay Mehrotra, Chairman and CEO, Micron Technology

Long-term contracts with customers are turning into a hallmark of the sector. SK hynix reported that it has received long-term supply agreements with approximately 10 major patrons, thus ensuring stability in profits while solidifying its place in the market of AI servers.

HBM4 is the next fight

The next battlefield is HBM4, the latest high-bandwidth memory technology designed for AI accelerators.

Samsung kicked off its mass production of HBM4 last February. The company has achieved transfer speeds of 11.7 Gbps, which is far higher than the JEDEC’s minimum speed of 8 Gbps. Additionally, in Q2 of this year, the company confirmed that it had successfully delivered its first samples of HBM4E technology to major clients.

SK hynix has also kicked off its mass shipment of HBM4 and wants to increase its output in the second half of the year; meanwhile, Micron is making progress with the development of its own HBM4 products for upcoming GPUs dedicated to AI applications.

In addition to faster speeds, HBM4 also doubles memory channels from 16 to 32, which means improved bandwidth and enhanced power efficiency in line with increased demand for bigger and more complex AI models.

There is quite a significant commercial potential for HBM4. TrendForce predicts that Samsung will earn over $1.2 billion from the HBM4 sales this year as cloud providers are quickening the development of AI facilities.

The squeeze may reach phones and PCs

The AI boost comes with challenges. As manufacturers focus on producing more AI memory, there will be less room for producing traditional chips. Samsung predicts memory shortage will persist in the second half of the year despite increased production, as demand continues to keep a strong pace for server DRAMs, enterprise SSDs, and HBM.

According to TrendForce, the DRAM contract prices are expected to increase by 58% to 63% in the third quarter, whereas the NAND flash prices can rise by 70% to 75%. If these forecasts prove accurate, it would mean some of the highest quarterly price increases in years as AI servers consume manufactured goods that were previously used for smartphones and PCs.

According to IDC, memory markets are predicted to stay structurally tight until 2027. It is estimated that under its downside scenario, global smartphone shipments will drop by 5.2% in case manufacturers transfer the extra costs of memory to consumers.

The combined results of Samsung, SK Hynix, and Micron indicate that the AI infrastructure surge is not just benefiting producers of GPUs (Graphics Processing Units) like Nvidia. Producers of memory products are benefiting the most as well from this boom, thus their profits are becoming a significant indicator of investments in enterprise AI globally.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.