SK Hynix, Samsung, Micron dominate AI memory

SK hynix, Samsung Electronics and Micron sit at the center of the AI memory market as demand for high-bandwidth memory, or HBM, outpaces broader semiconductor supply, according to company disclosures, analyst research and industry reports. HBM is the memory stacked next to AI processors from companies such as Nvidia, and it has become a critical input for training and inference systems. May 14 commentary circulating among investors described memory as a chokepoint for generative AI and argued that the three suppliers now have unusual leverage over the economics of large AI deployments. That framing matches the structure of the market: TrendForce said in a first-quarter 2026 HBM industry analysis that SK hynix maintained its lead, Samsung was recovering through HBM3e and HBM4, and Micron was expanding aggressively. (news.skhynix.com) Meta, Microsoft and Amazon have all reported sharply higher AI-related spending or infrastructure demand in recent weeks, adding to pressure on the supply chain that feeds accelerator systems. Meta reported first-quarter 2026 revenue of $56.3 billion and net income of $26.8 billion on April 29, while Microsoft said its AI business surpassed a $37 billion annual revenue run rate and Amazon said rising property and equipment purchases reflected investments in artificial intelligence. (trendforce.com) ### Why does memory matter more than a standard chip shortage? HBM is not interchangeable with commodity DRAM, and AI systems need it in large quantities alongside advanced packaging and accelerators. Samsung said in March that its sixth-generation HBM4 was in mass production and designed for Nvidia’s Vera Rubin platform, showing how tightly memory roadmaps are tied to next-generation AI servers. (sec.gov) SK hynix said on April 22 that first-quarter revenue rose to 52.5763 trillion won and operating profit reached 37.6103 trillion won, both records, driven by sales of HBM, high-capacity server DRAM modules and enterprise SSDs. The company said strong AI infrastructure investment sustained demand even in what is typically a seasonally weak quarter. ### Why are SK hynix, Samsung and Micron the names investors keep focusing on? (news.samsung.com) TrendForce’s first-quarter 2026 HBM report identified the same three companies as the suppliers shaping output, pricing and market share in HBM. The report said SK hynix remained in front, Samsung was rebuilding share and Micron was increasing capacity, leaving a concentrated supply base for the memory used in AI accelerators. Micron’s role has grown as HBM demand has broadened beyond a single customer set. (news.skhynix.com) While the company’s current-year detailed supply disclosures were not fully available in the sources reviewed here, Micron has publicly positioned HBM3E and HBM4 as core AI products, and industry reporting has repeatedly tied it to the same constrained supply chain as SK hynix and Samsung. (trendforce.com) ### How big is the spending wave behind this demand? Goldman Sachs Research said in a December 2025 note that consensus estimates for 2026 capital spending by AI hyperscalers had risen to $527 billion from $465 billion earlier in the earnings season. That figure is close to the upper end of the $300 billion to $500 billion range cited in May 14 market commentary about AI infrastructure buildouts through 2027. (investors.micron.com) Microsoft said on April 29 that quarterly capital expenditures and finance leases were $31.9 billion, according to CNBC’s summary of the earnings report, and that its AI business had surpassed a $37 billion annual revenue run rate. Amazon said trailing-12-month free cash flow fell to $1.2 billion, driven primarily by a $59.3 billion year-over-year increase in property and equipment purchases that “primarily reflects investments in artificial intelligence.” (goldmansachs.com) ### Where does Meta fit into the comparison investors are making? Meta reported April 29 first-quarter net income of $26.773 billion and diluted earnings per share of $10.44, helped by an $8.03 billion income tax benefit, according to its filing and earnings release. That gives investors a visible earnings benchmark when they compare the profit pool in AI infrastructure with the companies supplying the parts required to build it. (cnbc.com) The May 14 commentary’s claim that memory suppliers could command a value equal to two to three times Meta’s expected 2026 earnings is an investor argument, not a company forecast. What can be verified is that Meta, like its peers, is increasing AI investment while memory makers are reporting record or near-record results tied directly to AI demand. ### What should readers watch next? (sec.gov) Samsung has tied its HBM4 roadmap to Nvidia’s Vera Rubin platform, and SK hynix has said it will strengthen performance, yield, quality and supply stability in HBM. Those product ramps will be a direct test of whether supply can keep pace with hyperscaler and accelerator demand in the second half of 2026. (sec.gov) The next concrete checkpoints are company earnings and product updates from Meta, Microsoft, Amazon, SK hynix, Samsung and Micron, along with market-share estimates from firms such as TrendForce. Those disclosures will show whether HBM supply remains the binding constraint on AI system deployments through 2026 and into 2027. (trendforce.com) (news.samsung.com)