CXMT vs SK Hynix vs Samsung: Who Controls AI Memory Supply in 2026

SK Hynix holds the AI-memory supply crown with HBM demand exceeding its output by more than 2x and orders booked through 2029, while Samsung aggressively ramps standard DRAM and CXMT holds just 9% of DRAM revenue with HBM3E still in pre-production.

The number that settles the AI-memory ranking is SK Hynix's own disclosure: customers are demanding more than twice the HBM it can produce, with its DRAM and HBM order books filled through 2029. Against that backdrop, Samsung is ramping standard DRAM aggressively enough to have driven CXMT's Q3 2026 revenue share down to 9%—yet Samsung's own senior executives had to receive Sam Altman in Seoul, with Altman explicitly seeking more RAM for data centers amid what multiple reports describe as crushing shortages. CXMT, meanwhile, reported 613% Q3 revenue growth and returned to profitability in its nine-month 2026 earnings—sales growing more than sevenfold in absolute terms per one analyst's reading of that filing—but that growth is concentrated in commodity DRAM tiers, not in the HBM segment that now sets prices and margins across the entire memory complex.

On standard DRAM capacity, Samsung is the aggressor. Research cited by Wccftech attributes CXMT's quarter-over-quarter share compression directly to Samsung's ramp, and the implied dynamic is a price-taking environment for CXMT: growing revenue fast through volume and a rapidly maturing process, but losing relative position to a larger incumbent. CXMT's 4F²-based sixth-generation DRAM platform was validated per multiple October 2026 reports and targets DDR5 server samples by year-end—4F² cell geometry is the architecture the entire industry must eventually reach—but the company's own language is deliberate: validated prototype, gradual mass production, year-end samples. SK Hynix is not competing on commodity DRAM economics right now; it has effectively rationed its standard DRAM alongside HBM, with both sold out through 2029, harvesting margin rather than fighting for incremental wafer share.

The HBM gap is structural. SK Hynix is the established leader: some buyers are accepting 8-Hi configurations where higher-stack products were originally preferred, yet orders still top twice its supply capacity, and HBM prices are reported up 121% as that 8-Hi migration occurs. The practical implication is that SK Hynix is repricing its product mix upward while simultaneously being supply-constrained—a position of exceptional leverage. Samsung is in the HBM market and confirmed its relevance when Altman met roughly 50 Samsung executives in late October explicitly to secure memory supply; grey-market data is consistent with this, with smuggled HBM reportedly fetching $4,000 per unit as Huawei seeks product from both Samsung and SK Hynix. CXMT is in a different category entirely: it is reportedly preparing HBM3E volume production for AI hardware, but no confirmed mass-production shipments appear in its Q3 earnings disclosures, and its near-term technology roadmap centers on DDR5 server DRAM samples, not shipping HBM stacks.

Pricing power maps directly onto that hierarchy. SK Hynix forecasts memory shortages persisting into 2030, and its profit reportedly soared to $62 billion according to Korea JoongAng Daily—a figure that, whatever its precise scope, underlines how profoundly AI-driven demand has reset memory economics. Samsung's memory segment is separately described by industry reporting as virtually printing money, and Apple alone reportedly faces a $33 billion manufacturing cost increase as Samsung and peers prioritize AI server memory over consumer segments. CXMT operates in the commodity layer of this market: its 9% share and absence from HBM production mean it absorbs pricing dynamics set by others. Its doubled share buyback and resulting double-digit stock surge reflect domestic investor confidence and healthy cash generation from revenue gains, but buyback scale is a consequence of pricing power, not a source of it. The longer-dated leverage play is what analyst Dan Niles is flagging: CXMT reportedly targets 20% DRAM revenue share by 2028, a level that would force a structural response from both Micron and Samsung.

CXMT's path to changing the ranking runs through two specific execution tests. First: HBM3E volume production. If CXMT delivers confirmed shipments to Chinese AI hardware customers before mid-2027, it enters the AI-memory supply calculus for the first time and acquires the kind of leverage—captured demand, sticky qualification cycles—that SK Hynix currently holds in the West. Second: 4F² at scale. Sample validation is not wafer yield; moving from engineering samples to economically competitive mass production is the step that separates technology ambition from market-share gain. The G5 product flagged by XenoSpectrum as a potential picture-changer remains unquantified in available coverage and should be treated with appropriate caution until specifics emerge. On the established-player side, SK Hynix has formed a $10 billion U.S. investment unit and its AI arm has made material downstream acquisitions, moves that signal deliberate integration of supply-chain geography with AI-system distribution—a moat neither Samsung nor CXMT has built in the AI-infrastructure layer.

Three signals will determine whether the current ranking holds. First, CXMT's Q4 2026 DRAM revenue share: recovery toward 10-11% despite Samsung's ramp would validate the 20%-by-2028 thesis; further slippage would suggest the company is growing revenue on volume alone in a structurally price-down market. Second, Samsung's HBM customer-qualification disclosures: until Samsung announces design wins comparable in scale to SK Hynix's sold-out-through-2029 posture, the AI-memory tier gap remains. Third, the timing of SK Hynix's first confirmed HBM4 production ramp: XenoSpectrum positions HBM4 as a looming inflection, and whichever supplier ships HBM4 at volume first will set the pricing anchor for the next AI-infrastructure refresh cycle. As of November 2026, SK Hynix holds the AI-memory supply crown by a margin no plausible near-term scenario closes; Samsung is a strengthening second with DRAM momentum and HBM presence; CXMT is a credible DRAM challenger and an HBM aspirant whose execution record is not yet established.