Devoured - May 01, 2026
AI Has Made Memory Chips One of the World's Most Profitable Products (8 minute read)

AI Has Made Memory Chips One of the World's Most Profitable Products (8 minute read)

AI Read original

Memory chip makers are posting record-breaking profits as AI demand pushes Samsung, SK Hynix, and Micron into the ranks of the world's most profitable companies.

What: Samsung reported first-quarter 2026 net profit exceeding $30 billion (94% from semiconductors), with the three major memory chip makers collectively projected to generate roughly $350 billion in net profit for the year—vaulting Samsung past Alphabet, Microsoft, and Apple to become the world's second-most profitable company.
Why it matters: The profit explosion stems from a supply crunch where memory makers prioritized specialized high-bandwidth memory (HBM) for AI training, constraining conventional memory chips, while inference workloads for deployed AI models are now driving additional demand for general-purpose server memory—creating a "super boom cycle" expected to intensify next year.
Deep dive
  • Memory prices in Q1 2026 grew nearly 100% quarter-over-quarter, roughly double the initially projected 50% increase, according to TrendForce
  • Samsung's Q1 2026 net profit of $30 billion exceeded not only its prior quarterly record but nearly matched its historical high for an entire year
  • The three memory chip makers (Samsung 36% market share, SK Hynix 32%, Micron 22% for DRAM) are expected to rank among the world's top 10 most profitable companies in 2026—none cracked the top 10 a year ago
  • Samsung shares have risen 72% since the start of 2026, SK Hynix up 90%, and Micron up 65%
  • The supply crunch is expected to worsen in 2027, with Samsung stating "available supply is far short of customer demand" based on prebooked orders
  • The profit surge follows a two-phase demand pattern: first, specialized HBM production for AI training (paired with Nvidia GPUs) constrained conventional memory supply
  • Second, inference workloads for deployed AI models sparked additional demand for general servers using conventional DRAM and NAND flash memory
  • The three companies collectively control the overwhelming majority of both DRAM (90%) and NAND flash (55%) markets
  • While questions persist about whether AI services will generate commensurate profits, infrastructure providers are capturing an "epic windfall"
  • Memory makers gave priority to HBM production over conventional chips used in smartphones, PCs, and general servers, creating the supply constraint that drove prices up
Decoder
  • HBM (High-Bandwidth Memory): Specialized memory chips designed for AI training workloads, typically paired with Nvidia GPUs for training large language models
  • DRAM: Dynamic random-access memory, the main volatile memory used in computers and servers for active tasks
  • NAND flash: Non-volatile memory used for storage in SSDs, smartphones, and data centers
  • Inference: The phase of AI computing where trained models respond to user queries, as opposed to training new models
  • LLM (Large Language Model): AI models like GPT that require massive memory during training
Original article

The AI boom has pushed the memory-chip industry into a super boom cycle with record-smashing profits. Samsung has reported first-quarter net profit equivalent to more than $30 billion, blowing away its prior quarterly record and almost topping the company's high for full-year profit. The historic run doesn't look likely to end soon. The supply crunch is expected to grow worse next year.