Even if Interest Rates Suppress AI Stock Prices, Demand Cannot Be Broken…The Truth of Semiconductor Shortage

Key market variables currently include the Treasury's bond buyback policy and efforts to stabilize long-term interest rates. While the Treasury aims to stabilize rates through bond buybacks, some argue this method alone is insufficient to provide adequate liquidity. Thus, there is a suggestion for additional liquidity measures, such as stablecoins.

Those taking short positions in the current market are concerned about two main risks. First is the AI industry's monetization risk, with uncertainty surrounding actual profit generation compared to massive initial investment costs. Second is the US fiscal risk, fearing the negative impact of worsening finances on the overall economy.

Alphabet, for example, saw its stock price fall when its cash flow turned negative. However, Amazon states it can recover AI server investments and generate profit within three years. Since most AI customer contracts are for five years, Amazon is alleviating concerns that AI CAPEX will be prolonged. This trend shows big tech companies view AI investment not just as a cost, but as a strategic investment with a short-term recovery potential.

Despite having significant debt, CoreWeave is investing approximately $35 billion to $40 billion in CAPEX (capital expenditure). More notably, as of July, it has already secured an overwhelming backlog of $120 billion. This figure is more than three times the planned CAPEX for next year, clearly demonstrating that AI-related demand far exceeds supply.

In the AI industry, more crucial than the pace of CAPEX increase is the actual utilization rate of data centers. The key factor determining this utilization rate is 'token growth rate,' and if agent AI-driven token growth rate falls below 11-12 times annually, monetization issues may arise. Currently, major AI companies like Google and Anthropic are showing annual token growth rates of 15-20 times, leading big tech companies to continue AI infrastructure investments with the expectation that such high growth rates will be sustained.

The memory capacity required each time an AI agent is executed is rapidly increasing year by year. Specifically, demand for KV cache memory is predicted to increase by 3 to 6 times in the next two years compared to this year. Given this clear basis for calculating increased memory demand, big tech companies have no choice but to continue AI-related capital expenditures, even during the current period of rising interest rates.

Big tech companies' AI investments will continue as long as the need for token production persists. They are making aggressive investments because they foresee a potential 15-20 fold increase in token demand annually. Therefore, even if interest rates rise one or two times, the big tech's resolve for AI infrastructure investment is unlikely to waver, except for weak links. Investors currently taking short positions might be overlooking this explosive increase in token demand and the importance of the AI memory calculation structure.

Agent AI has been introduced for about two years. Therefore, if semiconductor supply becomes excessive when the learning curve plateaus, problems could arise. However, a decline in the utility of Agent AI is currently difficult to foresee. The possibility of a slowdown in token demand growth due to technical limitations is expected to become apparent only after at least 3-4 years. This means there is no need to worry about a slowdown in AI demand in the short term.

Although Samsung Electronics, SK Hynix, Micron, and even China's Changxin Memory are entering HBM (High Bandwidth Memory) production, there are physical limits to expanding supply. Due to wafer yield loss (a phenomenon where wafers split into three) during the HBM manufacturing process, production cannot be easily increased. This structural constraint means the supply side of HBM, i.e., the supply volume, cannot rapidly increase, indicating that the HBM shortage will continue.

U.S. mutual funds are not reducing their AI exposure; instead, they are maintaining it. They are adopting a strategy of diversifying their portfolios across the entire AI ecosystem, not just investing in AI semiconductor companies, but also in big tech, optical communications, power, and nuclear energy. This suggests that they highly value the growth potential of the AI industry and are making long-term investments.

If the Democratic Party secures a majority in the midterm elections, AI investment or EV-related policies discussed so far could unfold in entirely different directions. Therefore, up to the first half of October, the midterm election issue, and from now until October, the direction of interest rates are expected to be key market variables.

The current market is experiencing reduced liquidity due to a decrease in customer deposits. As a result, sector rotation between large-cap stocks like Samsung Electronics and other sectors is inevitable. When Samsung Electronics shows strength, other stocks may relatively weaken or consolidate, and when Samsung Electronics takes a breather, other stocks may come into focus. Therefore, investors need a strategy to agilely respond to market changes by holding a mix of stocks with and without AI exposure.

NVIDIA's stock price historically rose before most earnings announcements for about three years since 2023. Even though actual earnings release days often saw surprises of 5-6% or exceeded expectations, the stock tended to fall after the announcement due to expectations already being priced in. This is interpreted as investors buying in anticipation of stock price increases before the earnings release.

The US Treasury's attempt to lower long-term interest rates through bond buybacks is being interpreted by some as QE5 (fifth round of quantitative easing), sparking controversy. This policy is important because it could be seen as the US signaling its intention to tolerate a weaker dollar through interest rate cuts. This, in turn, fuels market skepticism that it could lead to a decline in the value of the US dollar. Fed Chair Powell's remarks at the upcoming Jackson Hole meeting are expected to provide important clues for interpreting these policy intentions.

Samsung SDS's business structure will add an AI infrastructure division to its existing IT services, cloud, and logistics sectors. By 2030, an estimated ₩6-7 trillion in revenue is projected from AI data centers alone. This is expected to bring the total business value to the high ₩20 trillion range. The AI infrastructure business is analyzed to add approximately ₩8 trillion to the total corporate value, acting as a future growth engine for Samsung SDS.

Samsung SDS, also operating various businesses, tends to be discounted due to a lack of information on each segment. The long-term goal is to transparently disclose and properly value each business segment. The possibility of undervaluation compared to the current stock price of ₩220,000-₩230,000 is raised.

This involves building a system that assigns and controls robot missions in real-time according to production plans. If Samsung Electronics creates the robot hardware, SDS will be responsible for the overall conductor role as the operating software provider. To achieve this, it has made a strategic investment in and collaborated with 'Walden Robotics' in the US, building a business model linked with MES (Manufacturing Execution System).

There is ample potential to expand the business to 35 manufacturing sites within the Samsung Group, which could lead to potential operating profits of ₩2.5 trillion. This highlights the importance of the robot business as a new growth engine for Samsung SDS.