The short version
In mid-October, statements from the world's two biggest memory buyers read as a public pressure campaign. Apple hardware chief John Ternus told Time that memory constraints are 'absolutely a crisis' and that all options should be on the table — remarks widely read as leaving the door open to Chinese memory. Tesla's Elon Musk posted on October 13 that Micron had been 'reasonable regarding future memory prices, while others have been extortionate,' and on October 15 that China 'will be making a LOT of RAM in 2028.' The apparent message to Samsung and SK hynix: cut prices, or we buy Chinese. Whether the two companies coordinated is not established. None of the reports cited here describes a discount resulting from it. The reporting gives CXMT little reason to offer one: its DRAM pricing is put at or above Samsung's level, and its capacity is reportedly prioritised for Chinese customers such as Huawei, ByteDance, Tencent and Xiaomi, and Apple's engagement remains at the testing and early-talks stage (WSJ reporting, via MacRumors). Our read, consistent with Korean and Taiwan trade coverage: the China card is a negotiation posture, not a price lever, because Chinese memory is constrained on price, volume, and politics. For component buyers, this is the most instructive negotiation story of 2026. It shows exactly what does and does not create leverage in a seller's market — and the answer is not what the headlines suggest. This article breaks down the signals, why they aren't working, and the four sources of leverage that actually work.
The signals: two CEOs, one apparent message
The signals that read as a pressure campaign landed within a week in October: - **October 13–15 (Musk):** Tesla's CEO posted on X that Micron had been 'reasonable regarding future memory prices, while others have been extortionate,' followed by a post saying China 'will be making a LOT of RAM in 2028.' Analysts read the unnamed "others" as Samsung and SK hynix. - **October 15 (Ternus):** Apple hardware chief John Ternus told Time that memory constraints are 'absolutely a crisis' and that all options should be on the table — remarks widely read as leaving the door open to Chinese memory. - **Earlier (WSJ, August 9, via MacRumors August 10):** The Wall Street Journal reported that Apple has been testing CXMT DRAM across product lines including iPhones and MacBooks, and has held early talks about devices sold in China. The context is straightforward. As Samsung and SK hynix divert wafer capacity to HBM for AI data centers, conventional DRAM for phones and PCs has become scarce and expensive — Q4 contract prices are expected to rise another 10-15% for conventional DRAM and 15-20% for NAND (TrendForce). The two Korean suppliers hold the pricing power, and their biggest customers are trying to break it by brandishing a Chinese alternative.
Why no discount has surfaced in the reporting so far
The flaw is structural, and it predates the October statements. Apple's engagement with CXMT — testing across product lines (WSJ via MacRumors, August) and early talks about China-market devices — has produced no announced supply agreement, and the pricing reporting points the other way: multiple assessments put CXMT's DRAM pricing at roughly Samsung's level, sometimes slightly higher. A challenger in CXMT's reported position — capacity prioritised for home customers, prices at market level — may have little reason to discount for a prestigious customer who has not yet committed volume. That is our read of the cited reporting, not a disclosed CXMT position. Why would a challenger supplier hold at market price for the world's most prestigious customer? Three reasons, each a lesson in market structure: **1. CXMT's reported priority is its home customers.** CXMT is reportedly prioritising Chinese customers such as Huawei, ByteDance, Tencent and Xiaomi, with little capacity left for new international customers. A supplier whose capacity is committed to home customers first has little reason to discount for a prestigious customer who has not yet committed volume. **2. CXMT prices at market, not below it.** In a seller's market, every supplier charges what the market will bear. Multiple assessments put CXMT's DRAM pricing at roughly the same level as Samsung's — sometimes slightly higher. A new entrant with constrained capacity has no incentive to start a price war, and CXMT's July 27 listing on Shanghai's STAR Market (raising RMB 57.9 billion, about US$8.6 billion, per Reuters and CNBC) means it now has investors expecting returns. **3. The alternative isn't equivalent yet.** CXMT runs about 300,000 wafers/month of DRAM capacity across three 12-inch fabs (Hefei ×2, Beijing ×1, per TrendForce), with SemiAnalysis projecting 350,000 by end-2026, 420,000 by end-2027, and 500,000 by end-2028 — about 17% of global DRAM capacity in 2027 (SemiAnalysis). But wafer capacity is not revenue: CXMT held about 7.6% of global DRAM revenue in Q1 2026 and 9.5% in Q2 (TrendForce), analyst estimates carried in trade coverage put it roughly five to seven years behind Samsung (treat the figure as approximate), and SemiAnalysis estimates its DDR5 costs per unit of capacity are more than 30% higher (a cost estimate; the standard explanation is no EUV access and DUV multi-patterning). The capacity that matters for Apple-class products — high-end mobile DRAM at scale — is the capacity CXMT is shortest on.
The politics: a card you may not be allowed to play
Even if the economics worked, US buyers face a regulatory wall: - **CXMT is on the Pentagon's Chinese military companies list (Section 1260H).** Section 5949 bars federal agencies from procuring covered products from listed companies, and the senators' letter argues the practical risk runs further. What any given company may or may not do with a listed supplier is transaction-specific — a compliance question to clear with counsel, not a blanket prohibition we can state. - **Senators' letter (July 30):** In a July 30 letter, seven bipartisan senators led by Schumer and Banks asked Apple to commit not to use CXMT or YMTC memory in any product, arguing that once a part clears qualification, extending it worldwide is 'a single procurement decision away.' - **Commerce Secretary Howard Lutnick** was reported (via ChosunBiz's account of WSJ reporting) as saying it is not desirable for U.S. corporations to use Chinese-made memory — a policy signal reported second-hand, not a formal prohibition. - As of the August reporting (MacRumors), Apple was reportedly seeking White House sign-off before proceeding, with China-market devices the subject of early talks; no approval had been reported at that time. Korean and Taiwan trade coverage reaches the same conclusion: under the triple constraints of technology, cost, and regulation, the card cannot threaten the Korean duopoly in the short term — a trade-press verdict we share, not a measured result.
What this means for component buyers
You are not Apple. You cannot get a CEO interview, move a stock, or lobby a commerce secretary. But this episode reveals the mechanics of leverage in a seller's market, and the mechanics apply at every scale: **1. Leverage comes from committed alternatives, not mentioned ones.** Apple's problem was that its "alternative" was a press statement. CXMT could see that Apple's qualification, volume, and politics weren't real yet — and, we expect, priced accordingly. A credible alternative means a qualified second source with actual allocation. Until you have one, threats are theater and suppliers know it. **2. The suppliers who discount are the ones who need the business.** Nothing in the reporting shows CXMT turning Apple away. What it describes is a supplier whose capacity is reportedly prioritised for customers at home and whose prices sit at market level — conditions that make a discount for Apple unlikely without anyone having to refuse one. Samsung and SK hynix are in the mirror-image position: with HBM demand absorbing their capacity, the cited reporting gives them no evident reason to discount either. In 2026, discount leverage doesn't exist — the only negotiation is over allocation and terms. Spend your energy there: volume commitments, term length, and delivery priority beat price haggling. **3. LTAs are the leverage now.** Notice who got CXMT's capacity: the Chinese customers it prioritises (Huawei, ByteDance, Tencent, Xiaomi). The same pattern is playing out at SEMCO (#55), at Micron, and across the AI supply chain. If you want supply security, the instrument is the LTA — signed before you're desperate, not after. **4. Chinese memory is a real option for some buyers — with eyes open.** HP has reportedly used CXMT products in non-US markets. For products sold outside the US, without US-government exposure, Chinese DRAM/NAND can be a legitimate second source — subject to your own entity-, item- and transaction-level compliance review, since geography alone does not settle the question. Budget for parity pricing, qualify carefully (mind the technology gap), and document the regulatory analysis before committing. **5. Don't plan on a 2028 price crash.** Musk's 'China will be making a LOT of RAM in 2028' framing implies relief is coming. The capacity math says otherwise: CXMT's expansion (500K wafers/month by end-2028) arrives alongside Samsung/SK/Micron expansions, into a market where AI demand is still absorbing everything. New supply meets new demand. Plan for elevated prices through 2028, and treat any relief as upside. The same logic applies to the flash and memory-adjacent parts in your BOM. Winbond, GigaDevice, and Macronix NOR/NAND parts — the kind used for firmware storage in everything from routers to industrial controllers — are priced by the same seller's market. The buyers who secured them did it with forecasts and commitments, not threats. Whatever your memory exposure, the instrument is the same: qualify a second source now, commit volume early, and negotiate allocation instead of price.
Frequently asked questions
**Is Apple actually buying CXMT memory?** Not yet, based on reporting. Apple has been testing CXMT DRAM across product lines including iPhones and MacBooks (WSJ, August 9) and held early talks about devices sold in China. No supply agreement has been announced, and US political opposition is significant. Treat it as an option Apple is developing, not a done deal. **Why won't CXMT undercut Samsung to win Apple?** On the reporting cited here, it has little need to. Its capacity is reportedly prioritised for Chinese customers, its prices are already at market level, and discounting for Apple would anger existing customers and its new public investors. In a seller's market, challengers price at parity. **Will Chinese memory bring prices down in 2028?** Unlikely to crash them. CXMT's installed capacity is projected to reach ~500K wafers/month by end-2028 (SemiAnalysis estimates), but AI demand is growing too, and Chinese output is partly reserved for domestic customers. Expect gradual pressure at the margin, not a price collapse. **Can my company buy CXMT or YMTC memory?** Technically yes, through distributors and module makers — subject to your regulatory exposure, which depends on your entity, the items, and the transaction, not just where the end product is sold. Get compliance advice first. Prices will be at market, not below. **What should a small buyer do instead of playing the China card?** The unglamorous things that work: sign LTAs or commit forecasts early, qualify a second source (including Chinese vendors where legal), design for memory flexibility (multiple qualified footprints), and work through distributors with contracted allocation. Leverage is built, not announced. **Did the campaign achieve anything?** Possibly at the margins: it signals to Samsung and SK hynix that their largest customers are actively developing alternatives, which may moderate future increases or improve allocation for Apple and Tesla specifically. For everyone else, the lesson is the mechanics, not the outcome. ---
Sources
Use the latest guidance for the exact part and manufacturer when deciding acceptance criteria.
- U.S. Senate Committee on Foreign Relations: Shaheen, Banks, Schumer, Colleagues Demand Apple Reject Chinese Military-Linked Chips (30 Jul 2026)
- CNBC: Chipmaker CXMT's 466% market debut surge makes it the most valuable China-listed company (27 Jul 2026)
- TrendForce News: CXMT Reportedly Eyes Second Beijing 12-inch DRAM Fab; HP, Asus, and Acer Said to Begin Limited Use (4 Aug 2026)
- MacRumors: Apple Testing Chinese Memory Chips in iPhones and MacBooks (10 Aug 2026)
- ChosunBiz: Trump administration presses Apple to drop Chinese memory suppliers (15 Aug 2026)
- Seoul Economic Daily: AI Memory Boom Is Funding China's Chip Catch-Up (23 Oct 2026)