AI Speed Control Debate Increases Infrastructure Investment Uncertainty

Unlike existing transformer models that sequentially pass through 100 layers, loop transformers add a feedback (Loop) process in the middle of the layers. This is an efficient method that maximizes the depth of thought while reducing computing usage. It is considered a core structure for accelerating the achievement of AGI (Artificial General Intelligence).

Existing AI could output thought processes as text, allowing humans to identify and correct errors. However, loop transformers process thought processes in an internal latent space and do not disclose the process externally, leading to faster development. But this creates a structural limitation where human intervention or correction is difficult when problems arise.

The debate over AI development speed control is an emerging issue that can arise at any time, along with ensuring safety. At this stage, it is too early to judge that such controversy will have a significant impact on data center infrastructure investment. While it may negatively affect stock prices in the short term, mid- to long-term infrastructure investment is expected to continue.

Establishing independent organizations and the discussion process are expected to take considerable time. Investors should pay close attention to the performance and official comments of data center companies. Companies have not yet declared that they will delay investment due to safety concerns.

The current AI debate is interpreted as an attempt to bring the 'elephant in the room' to the surface and address it. This should not be seen as the resolution of uncertainty, but rather as a stage where uncertainty is revealed, leading to concrete discussions.

There is a possibility of unpredictable sell-offs, similar to the case in late July when Hynix's stock price plummeted almost to the lower limit in the morning. In such uncertain situations, it is necessary to wait until companies make clear statements.

The interest rate hike itself has already been reflected in market consensus, and the pace of future interest rate hikes and the final interest rate decision announced after the FOMC will be the market's main focus.

The market is now entering a phase of adapting to a high interest rate environment.

It is necessary to focus on sectors where earnings growth is supported after the interest rate hike period.

In the past, Hynix typically traded at a 20-50% discount compared to Micron. Assuming that Hynix and Samsung Electronics' earnings next year will decline by more than 10% due to the exchange rate, Hynix ADR is expected to trade at a PER of around mid-6s. While the disparity between the main stock and ADR still exists, it is evaluated that the worst situation has passed.

The expert emphasizes that they did not short the market, and as a shareholder of Samsung Electronics and Hynix, their purpose is to convey the market situation as it is. Since related trading markets are already reacting over the weekend, they clarify that their intention is not to create anxiety but to share a realistic analysis as an investor.

As the probability of interest rate hikes increases through CPI and PPI indicators, investors must acknowledge and respond to the 'new normal' of rising interest rates. Instead of vague anxiety, it is necessary to calmly observe the changing market situation with fundamental-based investment and an ROIC (Return on Invested Capital)-centric strategy.

A stronger won acts as a factor in the decline of earnings forecasts for the export-oriented Korean market. It is bound to affect major export industries such as Samsung Electronics, Hynix, automobiles, and shipbuilding. Even if the US monetary policy stops, uncertainty due to a stronger won is expected to persist.

The foldable phone market is still small, accounting for less than 2% of the total mobile phone market. With the timing of Apple's iPhone foldable phone launch overlapping with a surge in semiconductor memory prices, Apple had no choice but to adopt a strategic choice of selectively imposing memory on foldable phones and setting high prices, rather than focusing on other products. Attention should be paid to the on-device AI implementation method rather than just the device.

Apple is implementing on-device AI differently from PC AI. It adopts a method of having one core basic AI model and attaching specialized modules as needed. This appears to be a strategy based on realistic foundations. From a semiconductor perspective, D-RAM demand may be less than expected because the AI model is compressed, but by attaching AI functions in a modular fashion, data storage increases, and NAND flash memory may become more prominent.