AI Has Made Memory Chips One of the World’s Most Profitable Products
Samsung is expected to outearn Apple, Microsoft and Alphabet, while two other memory giants are projected to enter the top-10 profit list this year
SEOUL—At the end of last year, the world’s investments in artificial intelligence had already pushed the memory-chip industry into a “super boom cycle.” Profits smashed records. Prices in the opening three months of 2026 were expected to jump another 50% from the prior quarter.
But things didn’t go that way. They got better—a lot better.
On Thursday, Samsung 005930 -2.43%decrease; red down pointing triangle Electronics reported first-quarter net profit equivalent to more than $30 billion. That not only blew away its prior quarterly record, but also nearly topped the South Korean company’s high for full-year profit.
Around 94% of Samsung’s first-quarter operating profit came from its semiconductor business.
Samsung’s chief memory rivals—SK Hynix of South Korea and U.S.-based Micron Technology MU 2.81%increase; green up pointing triangle—recently delivered similarly dizzying results. Those three firms dominate the market for memory, which gets used alongside Nvidia’s processor chips for AI computing.
While fears grow over whether AI services will eventually reap big profits, an epic windfall is flowing to the companies involved in the build-out of related infrastructure.
The historic run doesn’t look likely to end soon. Based on Samsung’s prebooked orders, the supply crunch is expected to grow worse next year, said Jaejune Kim, the company’s executive vice president for memory. “The available supply is far short of customer demand,” he said on a Thursday earnings call.
Since the start of this year, shares of Samsung have risen by 72%. SK Hynix’s shares are up 90% and Micron has gained 65%.
Memory prices in the first three months of 2026 grew nearly 100% from the prior quarter, roughly double the initially projected growth, according to Trendforce, a tech-market researcher.
In recent years, memory makers had given priority to production of specialized memory needed for AI—called high-bandwidth memory or HBM—which in turn constrained the supply of more conventional memory chips used in smartphones, personal computers and general servers. Training large-language models typically requires the pairing of Nvidia’s graphics processing units with HBM.
More recently, demand has risen for inference, the type of computing that allows trained AI models to respond to user queries. That sparked more demand for general servers, which use conventional memory, and elevated the earnings of Samsung, SK Hynix and Micron into a new echelon.
The three companies are collectively expected to generate net profit of roughly $350 billion for 2026, according to FactSet estimates. Each is projected to rank among the 10 most profitable listed companies in the world—with Samsung vaulting past Alphabet GOOGL 0.05%increase; green up pointing triangle, Microsoft and Apple to the No. 2 spot. A year ago, none of the memory makers cracked the top 10.
A single chip-making factory, or fab, can cost upward of $20 billion and take years to build. Samsung, SK Hynix and Micron are building new facilities, but production won’t likely hit full levels until late 2027 or 2028, industry analysts say. Many production lines are allocated to HBM, which takes up more capacity compared with conventional memory.
Memory comes in two major types: DRAM, which provides temporary storage for faster data-processing in servers, PCs and other electronics, and NAND flash, which provides long-term data storage, such as storing photos on a phone.
HBM is made by stacking DRAM chips on top of one another, which is later packaged alongside processors made by the likes of Nvidia to speed up AI computing. Nvidia works closely with Samsung, SK Hynix and Micron.
Operating profit margins for both types have roughly doubled from usual levels, hitting around 80% for DRAM and up to 60% for NAND flash, said MS Hwang, a semiconductor research analyst at Counterpoint.
Many big companies in the server, PC and smartphone businesses are paying premiums to buy up bigger volumes of memory and constrain supply for rivals, Hwang added. “It’s the idea that whoever dominates memory supply can dominate AI,” he said.
“What we’re seeing today is the most severe memory shortage that the market has ever seen,” said Marcus Chen, executive vice president at FusionWorldwide, a global distributor of electronic components. Most of the customers Chen supports are getting only 30% to 50% of the memory chips they need. “Some even less,” he said.
Customers and memory-chip makers long relied on handshake deals for long-term supply, but are now moving to binding contracts in some cases. Some contracts go as long as five years and call on customers to prepay some 30% of the cost or share in the investment costs for new memory-chip factories, said Peter Lee, a semiconductor analyst at Citi.
“We’re seeing customers go this far,” Lee said.