Even if the development speed of frontier AI models slows down, the inferential AI market will not stop growing, ensuring continued demand for related semiconductors.
AI development should be understood as a single process, like a digestive system, and it is difficult to determine the overall situation based on fragmented statements alone. The advancement of AI technology is interconnected like an organism, making it challenging to isolate specific parts and adjust their speed.
This is interpreted as an effort to gain time to secure 'guardrails' as safety measures while slowing down AI development. Currently, AI model development is in the roadmap update phase from version 4.7 to 4.8, with technical preparations underway for the next version.
The U.S. government shows its commitment to AI regulation, with major companies like OpenAI, Anthropic, and Google discussing self-regulation. However, even if this regulatory framework is successfully established, the participation of rival China is pointed out as a realistic limitation. If China does not join self-regulation, the effectiveness of the regulation will decrease, potentially altering the international AI development competition.
In the current AI market, the proportion of training and inference is becoming equal, leading to a structural increase in demand for computing resources such as GPUs, NPUs, and memory. Therefore, the impact of adjusting the development speed of frontier models on the overall demand for the AI semiconductor market is expected to be limited.
The recent call by AI model companies for speed adjustment means returning from the aggressive growth trajectory (45 degrees) to the original growth trajectory (35 degrees) anticipated at the beginning of the year. This is interpreted as a move to adjust excessive speed and promote sustainable development as safety issues (guardrails) in AI development have emerged. Thus, speed adjustment is not a slowdown in growth but a return to the reasonably anticipated growth path.
Even though demand to meet 45-degree growth has not yet exploded, the supply shortage to meet already anticipated demand is intensifying. Even giant corporations like Apple are struggling to secure volumes in semiconductor price negotiations, indicating that the overall semiconductor supply remains tight.
Some claims of an 'AI market apocalypse' are refuted by the lack of clear reasons for frontier models and the inferential AI market to collapse. AI-related stock prices are based on productivity improvements, and the growth of the AI market acts as a key driver for enhancing productivity across the macro economy. Therefore, it is analyzed that if an AI market apocalypse were to materialize, it would lead to an unprecedented macroeconomic shock.
Elon Musk's mention of the 4.9 model suggests an Astra-level goal, and the 2.5 trillion parameter model roadmap remains valid. Big tech companies mentioning AI development speed adjustment is not an intention to stop performance competition but rather a move to gain technological leadership. They are analyzed to be employing a strategy to secure technological dominance by preempting regulations themselves.
It takes time even if there are meetings and whatnot for performance competition that has already started. So, it's not that easy. It's interesting.
While AI infrastructure demand itself will not disappear, there is a timing risk for investment in the short term. Companies may delay CAPEX (capital expenditure) execution due to earnings pressure. This could cause short-term investment timing uncertainty in the next six months starting from Q3, potentially affecting the performance of related companies.
The growth of AI agents will increase computing intensity, driving AI infrastructure expansion in the mid to long term. The recursive self-improvement process, where AI learns and improves on its own, generates enormous infrastructure demand. Due to these self-improving characteristics, AI infrastructure is expected to expand further in the long term, becoming a growth engine for related industries.
They aim to solidify their dominance by designing regulatory frameworks themselves before latecomers enter the market. Specifically, there is a strong strategic intent to defend recursive self-improvement performance competition from external competitors and secure technological leadership before the competition fully intensifies.
The characteristic of AI development, where human intervention decreases and AI evolves on its own, acts as a technological moat against competitors like China. This recursive computation demonstrates the enormous potential of AI computing demand and is expected to continuously drive demand for high-performance semiconductors.
Due to the nature of the HBM production process, internal capabilities must be concentrated, so OSAT (Outsourced Semiconductor Assembly and Test) companies like SFA Semiconductor and Hana Micron are expected to benefit. Especially in conjunction with the expansion of wafer packaging roles at Samsung's Onyang Campus, a positive trickle-down effect is anticipated across the entire post-processing industry.
SFA Semiconductor is expected to return to profitability next year due to a surge in DDR5 outsourcing test volumes at its Philippine subsidiary and expanded volumes including FPGAs (Field-Programmable Gate Arrays). Stable volume securing is anticipated through role sharing and certification completion with Samsung Electronics' own Vietnam factory, which will positively impact SFA Semiconductor's performance turnaround. Consistent performance improvement is expected in conjunction with the overall growth of the semiconductor post-processing market.
Doosan Fuel Cell's stock price maintains an upward trend with increased trading volume from the bottom. While current valuation is difficult to justify, analysts' target prices are trending upward from 34,000 won to 74,000 won. An investment strategy of applying a discount rate to the average target price is suggested, which is an investment approach considering long-term growth potential rather than short-term price fluctuations.
Fuel cells can be installed directly next to data centers, reducing transmission grid burden, offering fast power supply, and ease of installation. They contribute to solving carbon emission problems but have a limitation of small power scale. In a hybrid structure where nuclear power supplies large-scale power and LNG complements medium-scale, fuel cells are expected to play a crucial role in distributed generation.
Future power generation will shift from a single power source to a hybrid structure of various energy sources. In the long term, it is expected to move towards a combination of nuclear power, renewable energy, and ESS (Energy Storage Systems), and in short-term power shortage situations, a complementary hybrid model between power sources is essential. LNG gas turbines are efficient but still have carbon emission issues, and fuel cells offer flexibility that can be expanded according to required capacity, playing an important role in this hybrid structure.
Doosan Fuel Cell's primary PAFC (Phosphoric Acid Fuel Cell) technology uses phosphoric acid electrolyte, offering the advantage of being a mature technology that is easy to produce. In contrast, SOFC (Solid Oxide Fuel Cell) is much more advantageous in terms of electrical efficiency and points the direction for technological development. SOFC is still in the stage of technological completeness and verification, but considering long-term efficiency, technological development should move towards SOFC. In terms of short-term orders, PAFC is in a favorable position, with SK E&C and Miko actively involved in the SOFC field.
Answers come from the transcript, with the exact spot cited.
Want the next video from 삼프로TV 3PROTV too?
When 삼프로TV 3PROTV publishes, we'll write it up like the one you just read and email it to you.
삼프로TV 3PROTV published 99 in the last 7 days.
We skip Shorts. You can unfollow any time.
