Samsung’s A.I. Moment Is Here, but Is It Ready?
The introduction of ChatGPT has lit a fire under the shares of companies that produce microchips, the brains of artificial intelligence. Bets on the potential of so-called generative A.I. have poured in. The most eye-catching example of the rally is Silicon Valley’s Nvidia, the top seller of chips used in artificial intelligence, whose shares are up nearly 200 percent this year.
Samsung Electronics, the South Korean giant, is hoping to get in on the action. Widely known for its consumer products, Samsung also has the world’s largest memory chip business and the second-busiest semiconductor foundries, which build custom microchips for other companies.
Foreign investors have bought $8 billion worth of Samsung shares this year on the South Korean stock market — already the largest amount of foreign purchases in Samsung for any year since 2000, according to data provided by CLSA, an investment firm in Hong Kong. The surge reversed a sell-off over the previous three years, when foreign investors sold more of the company’s stock than they bought.
At an event in California last week, Samsung detailed what it called its “vision in the A.I. era.” Samsung believes it can snatch market share from the leading chip manufacturer, Taiwan Semiconductor Manufacturing Company, but recently the trend has gone the other way. According to Counterpoint Research, a market research firm, TSMC enjoys roughly 60 percent of total revenues in the global foundry business and Samsung only 13 percent — a gap that has widened since 2021 as some of Samsung’s customers, including Nvidia, have shifted their business to TSMC.
Samsung said it spent $7.4 billion in the first quarter of this year — when its profits fell a staggering 95 percent — on its chip business, a portion of which is expected to serve the A.I. industry. It is expanding production at its chip-manufacturing complex in Pyeongtaek, about 40 miles south of Seoul, as well as a chip factory in Texas. Over the next 20 years, Samsung said, it plans to work with the government on a $230 billion plan to build a chip-making “megacluster” in South Korea.
The optimism is tied to Samsung’s memory chip business, which makes up roughly half the company’s operating profit in an average year, said Sanjeev Rana, a senior analyst at CLSA.
Compared with traditional servers — the hardware that underpins desktops and databases — the servers built for artificial intelligence can require four times the memory, called DRAM. Samsung commands roughly 45 percent of the global DRAM market. And it is the only major memory company to invest in more production despite an industrywide tumble in memory prices, Mr. Rana added.
The chip industry is known for its boom-and-bust cycles. After a spike in demand for memory chips during the pandemic, chip makers began one of their worst downturns in years last fall. Samsung’s memory chip rivals, including Micron Technology in the United States and South Korea’s SK Hynix, said they would cut back on investments in production this year.
Some analysts think Samsung’s spending in the down cycle will pay off in the long run when the memory sector recovers, in part because of artificial intelligence.
“If demand comes back, they will be very ready,” Mr. Rana said.
But skeptics question whether Samsung can achieve the kind of indispensable role in generative A.I. that it has had in smartphones and high-resolution televisions. It lost out last year when Nvidia chose SK Hynix as its supplier for a high-powered memory chip expected to become a fast-growing business line because of its prominence in future A.I. servers.
SK Hynix controls roughly 50 percent of that market for high-bandwidth memory, or HBM, compared with Samsung’s 40 percent, according to TrendForce, a market research firm. Shares of SK Hynix are up more than 50 percent this year, surpassing Samsung’s gain of 30 percent.
Samsung said it had already begun supplying “key customers” with a competing version of HBM. The next generation of its HBM is set to launch this year, it added.
Samsung’s lag in HBM technology could be a symptom of broader issues, said Nam Hyung Kim, an analyst at Arete Research, an equity research firm. In a report in February, Mr. Kim wrote that Micron had also leapfrogged Samsung’s technology in DRAM and another type of memory, NAND flash.
“The problem with Samsung is they always want to be big,” Mr. Kim said. “They’re spending so much money, but they’re not the leader in technology anymore.”
Mr. Kim said Samsung should invest more in research and not worry so much about market share. “Samsung is a bigger player than Apple in smartphones,” he said. “But how many people think Samsung makes a better smartphone than Apple?”
Samsung said in a statement that it had been successful in several aspects of advanced semiconductor technologies and that it could offer customers “comprehensive solutions” in the evolving landscape of A.I. and other technologies.
Samsung’s own executives have offered a more sober diagnosis.
In May, the president of Samsung’s semiconductor division, Kyung Kye-hyun, acknowledged in a talk to university students that the company “lagged behind” TSMC by up to two years. The remarks, which circulated widely in Korean media, were a rare admission for a company that has long prided itself on its tech leadership.
Mr. Kyung went on to vow that Samsung’s memory chips would become a “core” of A.I. supercomputers by 2028. “We can outperform TSMC within five years,” he said.
Jin Yu Young contributed reporting.
Source: New York Times