Apple Can't Buy Memory at a Discount Anymore: CXMT, Sold Out, and the End of Supply Chain Leverage

The market has a story about Apple testing memory chips from China's ChangXin Memory Technologies. The Wall Street Journal called it an AI boom squeezing supplies. The headline frames a desperate device maker turning to a second-tier supplier because demand has overwhelmed the market.

The headline gets the symptom right and the cause wrong. Apple is testing CXMT chips not because demand has surged. It's because the three companies that produce 90 to 95 percent of the world's DRAM — Samsung, SK Hynix, and Micron — are sold out through 2026. Every wafer of HBM (high-bandwidth memory) they stack for an Nvidia GPU is three times the wafer capacity of a standard DDR5 chip, and hyperscalers have locked up their output through five-year-plus long-term agreements. The DRAM market isn't being driven by a consumer demand spike. It's being driven by a structural capacity reallocation that has left conventional memory buyers with no alternative but to lobby the Pentagon for permission to buy from a blacklisted Chinese company.

That is the inversion that changes the reader's judgment. This is not a demand story. It's a supply discipline story, and for the first time in Apple's history, the world's most powerful device buyer has no leverage in the memory market.

The consensus narrative

The prevailing interpretation is straightforward. AI data centers are consuming so much memory that the supply can't keep up, creating shortages that ripple into consumer devices. Apple, facing rising costs, is looking at Chinese alternatives. The shortage will ease when new capacity comes online.

That narrative is half-right and structurally incomplete. Yes, AI data centers are absorbing an extraordinary share of memory production — 70 percent of all global memory chips in 2026, up from 20 to 30 percent historically. But the constraint isn't running out of wafers. The constraint is that Samsung, SK Hynix, and Micron have deliberately reallocated their wafer capacity toward HBM, which generates three to five times the revenue per wafer as conventional DDR5. They are not struggling to meet demand. They are choosing where to direct it.

SK Hynix sold out its 2026 HBM allocation by mid-2025 and is already pricing for 2027. Samsung and Micron are sold out through 2026. Recent reporting indicates all three manufacturers have allocated their entire DRAM and HBM capacity through long-term agreements. The oligopoly has transitioned from a cyclical commodity market with single-digit margins to a structurally constrained industry operating at 50 percent-plus operating margins.

Why Apple is talking to CXMT

Apple began testing CXMT's DRAM chips for devices sold within China in July 2026, according to the Financial Times. Simultaneously, Apple has lobbied the Trump administration since late May or early June to secure broader government approval for CXMT products. The lobbying effort has drawn sharp pushback from U.S. lawmakers, with Representative John Moolenaar calling a potential partnership a "grave mistake". The Pentagon added CXMT to its 1260H blacklist — which flags companies with alleged ties to the Chinese People's Liberation Army — in late June.

CEO Tim Cook told the Wall Street Journal that price hikes were unavoidable due to the market-wide crunch and urged the government to reconsider restrictions, saying "everything needs to be on the table". Cook also told investors on an earnings call that the DRAM market primarily has three suppliers and that adding more would help the supply side.

What that statement reveals is how much Apple's position has changed. Historically, Apple could bully its suppliers into discounting on volume. That relationship has now changed. In August 2026, CXMT rejected Apple's request for a price reduction, quoting the same prices that Samsung and SK Hynix charge. CXMT does not need Apple's business. It already supplies Huawei and Xiaomi through multi-year contracts and is fully committed to domestic Chinese demand.

The implication is fairly straightforward. The world's largest device manufacturer is asking a state-backed Chinese startup on the Pentagon's blacklist for a discount, and the startup said no.

CXMT: the numbers behind the new player

CXMT is the world's fourth-largest DRAM producer. It held approximately 7.7 percent of the global DRAM market in 2025 and is projected to reach 15 percent by 2028. Its monthly wafer capacity tripled from approximately 100,000 at the start of 2024 to roughly 290,000 by the end of 2025. Citrini Research models it reaching 350,000 wafers per month by the end of 2026 and approximately 500,000 by 2028, which would represent roughly 17 percent of global DRAM supply.

CXMT completed an $8.6 billion IPO on Shanghai's STAR Market in July 2026 — Asia's largest offering of the year. Shares surged approximately 470 percent on the first day of trading, lifting its market value to approximately $487 billion. First-quarter 2026 revenue jumped 719 percent year-over-year to $7.5 billion. Full-year 2025 revenue was approximately $8 billion, up 130 percent from 2024, driven almost entirely by ASP increases in a supply-constrained market rather than market share gains over incumbents.

Table 1 below summarizes CXMT's trajectory against the incumbents.

Table 1. DRAM Market Structure (2026 Estimates)

Metric	Samsung	SK Hynix	Micron	CXMT
Global DRAM Share	~40%	~28%	~22%	~8%
HBM Allocation	Sold out 2026	Sold out 2026, pricing 2027	Sold out 2026	Minimal (~30K wafers/mo by end 2026)
Wafer Capacity (WSPM)	~500K+	~400K+	~350K+	~290K (2025), ~350K (2026E)
HBM Maturity	Yield-constrained on 12-layer	~80% yields, dominant	Credible second source	Struggling with HBM3 8-hi, ~25% estimated yield
Capex 2026	Aggressive	$30B+ across new fabs/packaging	$20B in mega-fabs	IPO-funded expansion

The table shows a bifurcated market. Samsung, SK Hynix, and Micron control the HBM stack and have allocated all their output. CXMT holds a growing but secondary position in conventional DRAM, with HBM production estimated at roughly 30,000 wafers per month by the end of 2026 — a fraction of its total capacity and years behind the leaders in yield and technology.

CXMT's technology node, currently at Gen 4 (equivalent to 1z-nano), lags the leaders by approximately three years. Its cost per bit for DDR5 is estimated at more than 30 percent higher than Samsung, SK Hynix, and Micron. CXMT's DRAM ASP is only 5 to 10 percent below leaders in Q1 2026, which means it is capturing pricing power from the market shortage, not from a structural cost advantage. When a company without a cost advantage can quote the same price as the industry leader, that is a sign the buyer has no alternatives — and the seller knows it.

The pricing signal

The numbers tell you how deep the shortage is. J.P. Morgan estimates DRAM prices have risen more than 400 percent from the start of 2024 to the end of 2026. DRAM spot prices surged approximately 700 percent year-over-year by July 2026. Korean DRAM export prices reached $64,000 per kilogram, up from under $11,000 a year earlier.

The ASP divergence between conventional DRAM and HBM in 2026 tells you which business is winning:

Table 2. 2026 ASP Projections by Segment (Source: Visible Visible Consensus via S&P Global)

Company	Conventional DRAM ASP Growth	HBM ASP Growth
Samsung	+116%	~8%
SK Hynix	+78%	~1%
Micron	+54%	~22%

Conventional DRAM prices are surging because supply has been deliberately constricted. HBM prices are growing modestly because all capacity is pre-sold under long-term agreements,capping upside on incremental volume. The manufacturers are getting paid whether they win or lose the HBM race. The buyers of conventional memory — Apple, PC makers, smartphone vendors, automotive OEMs — are absorbing the cost.

Micron's financials show what the oligopoly is doing with this pricing power. Q2 2026 revenue hit $23.86 billion, up 75 percent quarter-over-quarter and 167 percent year-over-year, with Q2 EPS of $12.20 versus consensus of $9.19. Gross margin reached 72.6 percent, operating margin 65.6 percent, and ROIC 58.6 percent. The stock has gained 207.5 percent year-to-date and more than 600 percent on a rolling annual basis, currently trading near $878. Free cash flow for the trailing twelve months was $26.2 billion, up 1,291 percent year-over-year, against capital expenditures of $25.3 billion.

Micron is not growing because unit demand for conventional memory has exploded. It is growing because every wafer not allocated to HBM commands a premium, and the total supply growth is capped at 16 percent for DRAM in 2026, well below the historical norm of 20 to 30 percent.

The timeline: when does this end?

The most important number in this story is not a price or a margin. It's the timeline. SK Hynix's CEO warned in July 2026 that the shortage will persist beyond 2030. Intel's CEO Lip-Bu Tan stated there will be no relief until 2028. IDC projects global DRAM supply growth of only 16 percent in 2026, with new fab capacity from Micron and SK Hynix not reaching volume production until 2027 at the earliest.

CXMT's capacity expansion is the wild card. Its projected 500,000 wafers per month by 2028 could represent 17 percent of global supply. But production lines launching in the second half of 2026 are unlikely to alter the global supply-demand balance until 2027 due to yield rate variables. And even if CXMT hits its targets, Samsung adviser Kyung Kye-hyun warned that aggressive Chinese capacity expansion could create a supply glut by late 2027 — which would reverse the current price rally but would not restore Apple's negotiating leverage, because by that point the Big Three will have built enough HBM capacity to continue prioritizing the AI stack.

The constraint migration is the point. The bottleneck has moved from cleanroom availability to long-term agreement allocation. Samsung, SK Hynix, and Micron are not deciding who gets memory based on price. They are deciding who gets memory based on who signed the five-year deal first.

Investor Takeaway

The key issue is not whether Apple will successfully qualify CXMT chips for China-bound devices. The more important question is whether the DRAM oligopoly maintains its supply discipline as Chinese capacity scales.

If Samsung, SK Hynix, and Micron hold the line on HBM allocation — as their capex commitments and sold-out LTAs suggest they will — the structural shortage in conventional memory persists. Apple faces higher bill-of-materials costs with no path to negotiation. The company's margin advantage, built on supply chain dominance over three decades, erodes in the one component category where it has the least leverage.

If CXMT and YMTC scale faster than projected, their capacity could flood the conventional DRAM market by 2028, reversing current pricing. But that scenario requires Chinese fabs to achieve yields approaching the Big Three's levels while operating without EUV lithography equipment — a constraint that adds years, not quarters, to any timeline.

The near-term implication is that memory costs for consumer electronics remain elevated through at least 2027. Apple's price hikes across iPads and Macs in 2026 were the first signal. Reports of potential $300 price increases on upcoming models are the second. For investors, the structural lesson is that the DRAM market has bifurcated into two separate games: HBM for AI infrastructure, where the Big Three are locked in under long-term contracts, and conventional DRAM, where buyers are pricing in a multi-year shortage with no guaranteed relief. Apple's CXMT testing is not a supply chain win. It's a signal that the buyer who once set the terms no longer has them.