The AI Boom Isn’t Breaking the Memory-Chip Slump—Yet
The ChatGPT-led craze is a big future driver of results, though not enough now as Samsung Electronics forecasts 96% drop in operating profits Samsung Electronics said revenue for the April-June period is expected to slide by 22% to 60 trillion won. Photo: PETER DASILVA/REUTERS By Jiyoung Sohn July 6, 2023 8:30 pm ET SEOUL—The current artificial-intelligence frenzy is expected to drive significant growth in memory chips. But that time isn’t now. The latest reminder came Friday, when Samsung Electronics, the world’s largest maker of memory chips, forecast a 96% drop in its operating profit for its second quarter—a sign that even the South Korean semiconductor giant isn’t able to capitalize on the moment. Memory plays an essential role in enabling more sophisticate
Samsung Electronics said revenue for the April-June period is expected to slide by 22% to 60 trillion won.
Photo: PETER DASILVA/REUTERS
SEOUL—The current artificial-intelligence frenzy is expected to drive significant growth in memory chips. But that time isn’t now.
The latest reminder came Friday, when Samsung Electronics, the world’s largest maker of memory chips, forecast a 96% drop in its operating profit for its second quarter—a sign that even the South Korean semiconductor giant isn’t able to capitalize on the moment.
Memory plays an essential role in enabling more sophisticated versions of AI-based online tools like OpenAI’s ChatGPT, which can generate humanlike responses from prompts. Working in tandem with graphic-processing units made by the likes of Nvidia, memory chips help speed up computing times—and are hence important for building faster and more complex AI applications.
But the initial bump from the rush into generative AI services isn’t yet enough to overcome a broader tech slump that has damped sales of semiconductors, including memory.
Samsung said it expects second-quarter operating profits of 600 billion won, or roughly the equivalent of $458 million, versus the prior-year period’s 14.1 trillion won. Revenue for the April-June period is expected to slide by 22% to 60 trillion won.
Chip maker Nvidia broke into the exclusive club of companies that have a $1 trillion market cap. WSJ’s Asa Fitch breaks down how Nvidia got there—and why AI is fueling the company’s rapid growth. Photo illustration: Annie Zhao
SK Hynix is projected to post an operating loss of 2.8 trillion won in the April-June quarter, with revenue for the period falling 53% on-year to 6 trillion won, according to FactSet.
Samsung and SK Hynix will announce their full earnings later this month.
On an earnings call last week, Sanjay Mehrotra, chief executive of Micron Technology, said generative AI tools like ChatGPT had driven higher-than-expected industry demand for memory and storage for AI servers in its most-recent quarter. But U.S.-based Micron, No. 3 in the memory business, reported a net loss of $1.9 billion and a 57% year-over-year drop in revenue for the quarter ended June 1.
This year, global revenue for all types of chips is projected to decline by around 10% to $515 billion, according to the World Semiconductor Trade Statistics. As a category, memory chips are poised for the worst decline among the major types of semiconductors, with revenue expected to drop by 35% to $84 billion, WSTS projects.
Prices of memory chips, more commoditized than other types of semiconductors, began falling steeply from the second half of last year and continued to slide through this year amid a supply glut. In the April-June quarter, contract prices of two major types of memory, DRAM and NAND flash, fell quarterly by 21% and 13%, respectively, according to projections by Bernstein Research.
Inflation and macroeconomic uncertainties have prompted sharp cutbacks in consumer and corporate spending on smartphones, PCs and servers, all of which drive chip sales.
But the future looks brighter. Generative AI will direct more sales to next-generation—and more profitable—forms of memory. And the worst of the current sales slump may soon be over, memory-industry executives have said, as customers resume purchasing after a period of inventory adjustments and cuts in production and investment help tilt supply-demand dynamics in their favor.
SHARE YOUR THOUGHTS
What is your outlook for Samsung? Join the conversation below.
Samsung, SK Hynix and Micron have all launched next-generation types of DRAM, a major type of memory, which is targeted for use in AI systems. Called “high bandwidth memory,” HBM stacks together multiple layers of DRAM as one, which can then be packaged as one unit with the graphic-processing units made by Nvidia
and others. That allows a larger amount of data to move between the memory and processor at once. The time it takes for data to move between the two chips is also reduced. That results in improved computing speed and efficiency.HBM is about five times more expensive than a standard DRAM chip, bringing bigger total profits to manufacturers, according to SemiAnalysis, a chip-industry consulting firm. Today, HBM accounts for less than 5% of worldwide memory revenue, but is expected to account for more than 20% of the total by 2026, SemiAnalysis projects.
SK Hynix is a leading HBM supplier to Nvidia and controls about half of the market as of last year, according to TrendForce, a Taiwan-based tech-market research firm. Samsung accounts for 40% of the HBM market and Micron has 10%.
SK Hynix expects its HBM revenue for 2023 to grow more than 50% year-over-year and projects further gains in the coming years, its chief financial officer, Kim Woo-hyun, said in an earnings call in April. Samsung too has pointed to the rise of generative AI applications like ChatGPT as a positive force for future memory demand.
The AI-related portion of worldwide DRAM revenue is projected to grow from 16% this year to 41% in 2025, analysts at Citi said in a recent note to investors.
Write to Jiyoung Sohn at [email protected]
What's Your Reaction?
