"AI Speed Bump" Rises, Shaking Semiconductor Stocks... Should You Sell Samsung Electronics and SK Hynix for Now?

◇Stocks Hit in AI Supply Chain

As the "AI speed bump theory" gains traction in the United States, attention is also focused on its potential impact on semiconductor-related stocks including Samsung Electronics, SK Hynix, and U.S.-based Micron. Market experts believe that semiconductor and AI supply chain-related stocks could face short-term setbacks. However, given that the long-term growth trajectory remains intact, they expect the fallout to be unlikely to persist for an extended period.

According to Bloomberg on the 13th (local time), market strategists pointed out that a cautious approach to AI development could trigger short-term selling pressure on semiconductor manufacturers and AI supply chain-related stocks. This is because concerns are growing that a cautious approach to developing advanced AI models could weigh on the earnings of related companies. Earlier, Dario Amodei, CEO of Anthropic, proposed slowing down AI development so that safety measures can keep pace with the advancement of AI capabilities, a stance publicly supported by OpenAI's Sam Altman and Elon Musk.

However, the prevailing view is that this latest setback will not derail the long-term growth trend of AI investments. Gary Tan, portfolio manager at AllSpring Global Investments, said, "While these remarks may lead to short-term stock price pressure, it is unlikely they will shake the long-term AI investment flow." He added, "Since AI development is still in its early stages, it remains questionable whether other players in the ecosystem would be satisfied with their current rankings and choose to slow down amidst rapid technological evolution."

There are also observations that the speed-bump theory could serve as an opportunity for companies to reduce infrastructure cost burdens and improve profitability. Billy Leung, investment strategist at GlobalX Management, noted, "It actually has the effect of extending the development timeline." He analyzed, "As AI commercialization and adoption continue to expand, if the pace of new technology development slows slightly, it will help transition from the infrastructure construction phase to a stage where revenue is generated using already established technologies."

Charu Channan, senior investment strategist at Saxo Markets, explained, "Demands for enhanced safety measures are expected to lead to new expenditures and expanded revenues in areas such as cybersecurity and AI monitoring. Even if development speeds moderate somewhat, demand for computing power will persist, which in turn further solidifies the long-term profit-generation structure of the AI industry."

Amidst this, OpenAI CEO Sam Altman stated in a Fortune interview that "given the situations unfolding regarding AI safety, now is not an appropriate time for an IPO," announcing that it would not pursue an initial public offering (IPO) this year. Anthropic CEO Dario Amodei also appeared on a U.S. broadcast, stating, "This does not mean we need to be terrified today or halt all development," but added, "It means that every generation of models we release must undergo proper validation."

◇Trump vs. Obama Jeon Hyeon (President) Divergent Stances= Major AI companies are reportedly discussing the establishment of a standard-setting body to create joint testing and auditing systems. IT specialty media The Information reported that since July, a working group composed of executives under the CEOs of Anthropic, OpenAI, and Google has been meeting regularly.

In this regard, Microsoft (MS) CEO Satya Nadella agreed with Amodei's call for AI speed control and the introduction of external oversight but emphasized that a small number of entities should not be responsible for AI monitoring and control. He stated, "It must have broad representation across the ecosystem, including academia, nations, and various sectors." On social media platform X (formerly Twitter), Nadella said, "Any journey toward superintelligence must be rooted in the core principle that 'if the AI we build does not benefit humanity or remains outside human control, it is not worth pursuing.'" He further proposed "human control" and "benefit sharing" as the Daejeon principles for superintelligence development.

Meanwhile, Barack Obama Jeon (President) stated at a closed-door fundraising event for the U.S. Democratic Party on the 10th that "if AI is not properly controlled, it could become dangerous," urging the Democratic Party to make AI response a key political agenda. In contrast, U.S. President Donald Trump drew a line against the speed-bump theory during his visit to Ireland on the 13th, saying, "There are too many negative forces highlighting things that should not be highlighted; they are focusing on events that will never happen." President Trump emphasized, "The United States is currently ahead of China in the AI field," and added, "The one who wins in AI becomes the ultimate winner."

◇Strong Demand to Continue for Now

Driven by expanded investment in AI (artificial intelligence) infrastructure, global server market sales have surpassed 200 trillion won on a quarterly basis for the first time. With increased adoption of high-performance components for AI computation, the average selling price (ASP) of servers also rose significantly. As demand has spread beyond HBM (high-bandwidth memory) to include DDR5 (double data rate 5) and eSSD (enterprise solid-state drive), all types of memory semiconductors (hereinafter "memory"), favorable market conditions are expected to continue for domestic memory companies such as Samsung Electronics and SK Hynix.

According to market research firm IDC on the 14th, global server market sales in the second quarter of this year reached $166.3 billion (approximately 223.49 trillion won), marking a record high on a quarterly basis. This is the first time a single quarter's sales have exceeded 200 trillion won. It represents a 32.7% increase compared to the previous record of $125.3 billion (approximately 168.47 trillion won) in the fourth quarter of last year.

Investments in AI infrastructure by hyperscalers (large-scale data center operators) and CSPs (major cloud service providers) drove market growth. In particular, while global server shipments increased by 15.4% year-on-year in the second quarter this year, sales revenue surged by 52%. This indicates that not only were more servers sold, but the ASP per server also rose significantly.

GPU (graphics processing unit) acceleration servers are a prime example. The ASP of GPU-accelerated servers rose 43.6% year-on-year to $170,200 (approximately 229 million won). Sales of other servers equipped with ASICs (application-specific integrated circuits) and similar components also surged by 237.6% during the same period. Analysts note that as AI infrastructure investment diversifies from being centered on NVIDIA GPUs to include proprietary AI accelerators developed by big tech companies, server specifications are becoming more advanced at a faster pace.

As the "value" of servers rises, Samsung Electronics and SK Hynix are expected to continue their growth momentum. This is because AI servers require not only HBM but also high-value, high-capacity memory such as DDR5 for servers supporting data processing and eSSD for storing large-scale data. The same context applies to comments made by NVIDIA CFO Colette Kress during a recent conference call announcing earnings for the second quarter of fiscal 2027, where she noted, "Today's memory shortages are largely caused by AI deployment itself."

In fact, rising memory prices are a key factor driving up server ASPs. IDC stated, "Server suppliers cite rising DRAM and NAND flash prices along with increased component allocation as the main reasons for price hikes." This is also why global big tech companies are entering into long-term supply agreements (LTAs) with memory manufacturers. Samsung Electronics has already signed LTAs with global big tech firms and has now entered negotiations with five additional major customers. Typically, a "rolling" contract method is applied, where the 5th-year agreement term is extended by one year each year. SK Hynix has also signed LTAs with over ten customers, including key clients.

With expanded AI infrastructure investment and strong memory prices, earnings estimates for both companies are rising. According to financial information provider FNGuide, the combined operating profit consensus estimate (average analyst forecast) for Samsung Electronics and SK Hynix this year stands at 656.054 trillion won, exceeding seven times their combined operating profit of 90.8074 trillion won last year.

An industry official said, "AI servers have higher memory capacity and product specifications than existing servers, creating a structure where growth in the server market directly leads to expanded demand for high-value memory." He added, "As long as AI investment continues, demand for server memory such as HBM, DDR5, and eSSD will remain robust."

◇AI Infrastructure Business Booming

LG Electronics is accelerating its efforts to capture the rapidly growing global AI data center cooling solution market. It is undergoing additional certification procedures for liquid cooling core equipment from NVIDIA. Based on this, the company aims to enhance its competitive edge in cooling solutions to address increased heat generation from high-performance AI servers and secure additional orders worth trillions of won.

According to the electronics industry on the 14th, LG Electronics is currently undergoing final quality certification from NVIDIA for its 1MW (megawatt) class "Coolant Distribution Unit (CDU)." It has reportedly conducted internal tests in line with performance standards required by NVIDIA and submitted relevant data.

LG Electronics received final quality certification from NVIDIA last July for its 600kW (kilowatt) class CDU. Following the 1MW certification, it plans to expand CDU certification capacity to 2.5MW and 4MW classes. Development of a 1.4MW CDU is virtually complete, with goals to launch 2.5MW and 4MW class products and begin certification procedures within this year.

CDUs are core equipment for liquid cooling in AI data centers. As performance of high-performance GPUs (graphics processing units) and HBM (high-bandwidth memory) has recently improved, the power consumption and heat generated by a single AI server rack are increasing rapidly. Existing air-cooling methods alone have limitations in handling the heat from high-performance AI servers.

Consequently, the importance of "Direct-to-Chip (DTC)" liquid cooling, which attaches cold plates to high-heat chips like GPUs and CPUs to directly absorb heat with coolant, is growing. In DTC systems, CDUs send coolant to servers, absorb heat from chips, collect the heated coolant, cool it down again, and circulate it back.

The larger the CDU capacity, the more heat it can handle. In particular, industry observers view 2.5MW class CDUs as a critical turning point in the next-generation AI data center cooling market. NVIDIA's next-generation "Vera Rubin NVL72" uses up to 330kW of power per server rack, which is about 2.7 times higher than previous generation models.

LG Electronics is expanding its thermal management business from its existing operations of cooling data center spaces using large chillers and HVAC equipment to include liquid cooling areas around server racks and chips. The strategy is to build a portfolio covering the entire data center cooling system, from chillers to CDUs and cold plates.

According to NVIDIA's AI Factory Infrastructure public information, only about 20 companies worldwide have received CDU certification. Passing this verification means meeting technical requirements such as cooling performance and stability demanded by NVIDIA AI servers. This forms the foundation for expanding orders to hyperscalers (large-scale data center operators) in the future.

Market growth is also steep. The global AI data center market is projected to grow at an annual rate of about 26%, reaching approximately $133.5 billion (about 185 trillion won) by 2034. LG Electronics is also expanding data center cooling system orders targeting global hyperscalers. Orders related to AI data center cooling in the first half of this year exceeded 600 billion won, with a goal of securing additional orders worth trillions of won by year-end.

An industry official said, "Since most global AI data centers currently use NVIDIA products, NVIDIA's certification carries significant meaning and its requirements are stringent." He added, "Server rack and data center design companies mostly seek to use certified products."