Investors known as 'Seohakgae-mi' (domestic investors who invest in overseas stocks) are recently focusing on Broadcom instead of NVIDIA, the leader in the American artificial intelligence (AI) semiconductor market. In the past week, domestic investors have net purchased more shares of Broadcom than any other overseas stock. This is interpreted as reflecting expectations for the customized semiconductors (eXtreme Processing Unit·XPU) that Broadcom is currently developing.
According to the Korea Securities Depository's securities information portal, Sebro, the overseas stock with the highest net purchase settlement amount from December 21-27 was Broadcom. During this period, domestic investors net purchased shares worth $12.377 million (approximately 1.823 billion won).
As of the 26th, the amount held by domestic investors in Broadcom was counted at $1.67981 billion. This marks an increase of 20.98% from the held amount a week prior ($1.38846 billion on the 19th). In other words, Seohakgae-mi bought about $300 million (approximately 429.4 billion won) worth of Broadcom shares over the week. The amount held in Broadcom has increased by 80.55% compared to a month ago.
As of the 26th, the amount held in AI semiconductor leader NVIDIA was $12.5127 billion. In terms of held amounts alone, this is significantly higher than Broadcom. However, the rate of increase has been slower. The held amount for NVIDIA has increased by 6.3% compared to a week ago ($11.76057 billion on the 19th).
Recent attention on Broadcom is due to the rise of customized semiconductors (ASIC) as an alternative to NVIDIA's graphics processing unit (GPU) in the AI semiconductor market. Tan Hock Eng, the CEO of Broadcom, recently noted during an earnings announcement that they are developing AI chips with three large cloud corporations. The three corporations mentioned by Tan are said to be Alphabet (Google), Meta, and ByteDance.
So far, NVIDIA has nearly monopolized the AI semiconductor market. NVIDIA's GPUs have been essential for generative AI services. Major tech corporations such as Microsoft, Alphabet, Amazon, and Meta have relied on NVIDIA's GPUs when building the data centers that serve as the core infrastructure for AI.
However, as Broadcom's customized semiconductor XPU emerges as an alternative, the atmosphere is changing. While NVIDIA's GPUs are highly versatile, they are expensive. In contrast, Broadcom's XPU is specifically manufactured to include only the necessary functions required by each corporation, leading to relatively significant cost reductions.
Expectations for tectonic shifts have also been reflected in stock prices. Broadcom's stock price on the 27th was $241.75, a 51.41% increase from a month ago ($159.67). Meanwhile, NVIDIA's stock price only increased by 1.23% during the same period.
Market experts expect the optimism surrounding customized semiconductors to continue for the time being. Moon Jun-ho, a researcher at Samsung Securities, said, "There are reports that custom semiconductor orders related to AI are coming in," adding, "Projects ordered this year will begin to show profits from next year."
Researcher Moon noted, "The networking environment of existing data centers also needs upgrades, and Broadcom is the top company in that market," stating, "We expect profit increases from these two aspects."
Analysts suggest that Broadcom is not only generating expectations but is also achieving actual growth. The sales related to Broadcom's customized semiconductors in the networking sector for the fourth quarter of this year have increased by 45.2% compared to the same period last year. Kim Seung-hyuk, a researcher at Kiwoom Securities, noted, "There were likely contracts related to custom semiconductor design with big tech companies."
Researcher Kim stated, "It's still early in chip design, but the increased profitability in the relevant sector implies that revenue recognition is progressing based on non-recurring engineering (NRE) contracts." He added, "It would be difficult to say that the stock price has risen solely due to expectations for the potential of customized semiconductors; actual growth must be taking place."