An illustration displaying the DeepSeek logo / Reuters-Yonhap

Samsung Electronics and SK hynix are expected to face growing uncertainties in the rapidly growing artificial intelligence (AI) memory chip business following the emergence of Chinese AI startup DeepSeek, which has dealt a blow to Nvidia and other U.S. AI tech firms, experts said Friday.

DeepSeek recently introduced its latest inference model, R1, which delivers performance comparable to OpenAI's o1. The startup said training the model cost just over $5.5 million despite using lower-grade Nvidia graphic processing units (GPUs) designed for the Chinese market.

While many details remain to be verified, this has raised questions about the competitiveness of major tech firms that have invested billions in AI infrastructure.

"The paradigm that large-scale investment is essential for AI growth has changed. Big tech companies are now likely to focus on more efficient strategies rather than blindly competing to secure the latest GPUs," iM Securities analyst Song Myung-sup said.

"This shift could negatively impact domestic memory semiconductor companies that have been supplying high-capacity high-bandwidth memory (HBM) to Nvidia in response to AI demand," he said.

In response to this recent development, investors have sold off shares of Nvidia, which holds a 70 percent share of the AI semiconductor market. On Jan. 27 (local time) alone, the company saw its market capitalization drop by approximately $589 billion, with no signs of recovery yet. Other U.S. stocks linked to its AI value chain are also facing struggles.

If demand for Nvidia's high-performance GPUs weakens as the market expects, SK hynix — which almost exclusively supplies HBM3e to Nvidia — could see its revenue decline significantly.

An illustration displaying the Nvidia logo and a decreasing stock graph / Reuters-Yonhap

HBM3e, the most advanced version of its kind, plays a critical role in enhancing GPU performance, and SK hynix generates 40 percent of its total DRAM sales from its HBM business. Samsung Electronics is also undergoing quality testing in preparation for supplying these chips to Nvidia, eyeing it as its next revenue source.

Reflecting the uncertainties, the stock prices of Samsung Electronics and SK hynix on Friday did little to ease market concerns. Samsung's shares dropped 2.4 percent to 52,400 won ($36.07), while SK hynix saw a sharp 9.8 percent decline, falling to 199,200 won. The decline came on Korea's first trading day since DeepSeek's announcement, following the country's Lunar New Year holiday.

Stocks related to power and electricity, which had previously surged on expectations of benefiting from AI infrastructure development, also dropped sharply.

"As we supply HBM for GPUs to multiple vendors, we are closely monitoring industry trends and considering various scenarios," Kim Jae-joon, executive vice president of the device solutions (DS) division at Samsung, told investors during a fourth-quarter earnings conference call on Friday.

"As both long-term opportunities and short-term risks coexist in the market, we will ensure a swift and timely response to its rapid changes," he said.

Beyond market dynamics, geopolitical tensions are adding another layer of uncertainty for Korean chipmakers.

To solidify AI leadership, the U.S. government is reportedly considering measures to block exports of Nvidia's low-end H20 chip to China, which is equipped with Samsung Electronics' HBM. In response, China may intensify support for its domestic semiconductor firms and accelerate its push for self-sufficiency, potentially reshaping demand for Korean chips.

"If the U.S. government uses this incident as grounds to impose a full-scale ban on the sale of Nvidia's China-specific GPUs, Korean DRAM manufacturers, which supply HBM for these GPUs, will also suffer negative repercussions," Song said.

The logo of SK hynix is seen on its product during the 26th Semiconductor Exhibition in Seoul, Oct. 23, 2024. Reuters-Yonhap

Strong global demand for Korean chips

Despite recent disruptions, industry experts see sustained growth ahead for Korean chipmakers. Many believe that Nvidia's market dominance will persist. After all, even DeepSeek's R1 was developed using Nvidia GPUs.

"If the same algorithm were applied to high-performance Nvidia chips, it could achieve even greater results than DeepSeek's," Lee Jong-hwan, a system semiconductor engineering professor at Sangmyung University, said.

Another potential upside is the technological advancements triggered by the DeepSeek breakthrough, which could contribute to the broader expansion of the AI ecosystem.

"As the market continues to grow, the AI semiconductor sector is poised for further expansion. The industry has now learned about the possibility that low-grade devices can deliver strong performance. Given that Samsung and SK also provide low-grade options, their HBM business could see further growth in this area," Lee said.

"HBM memory is absolutely necessary with the current technology, and Samsung and SK could see increased demand as a result," he said.

Daol Investment & Securities analyst Ko Young-min said, "There is still a need for technological advancements to achieve artificial general intelligence, and competition among big tech companies is expected to continue."

Ko continued, "Amid the ongoing trend of increasing capital expenditures for market dominance, the semiconductor and hardware sectors are likely to remain in a favorable environment."