CXMT's $8.6B Shanghai IPO and China's Memory-Chip Capital Cycle

TL;DR

CXMT lists on Shanghai's STAR Market on 27 July after raising $8.6 billion — Asia's largest IPO of 2026, 212× oversubscribed by retail, at an implied valuation of ¥579 billion (~$85 billion).
The company is now the world's fourth-largest DRAM maker with ~8% global share, up from ~3% a year ago. Q1 2026 revenue hit $7.5 billion — a 719% year-on-year surge.
It just signed a five-year, $7 billion+ supply deal with ByteDance and a $3 billion+ deal with Tencent. Apple is actively testing CXMT chips for China-market devices and lobbying Washington for clearance.
The IPO lands at the peak of a memory supercycle — DRAM contract prices rose 93–98% QoQ in Q1 2026 — but Michael Burry is short Micron, Morgan Stanley sees China supplying 30% of new global DRAM capacity by 2028, and Bloomberg Intelligence says the shortage peaked in Q2.
The capital cycle question: is CXMT raising at the top, or is this the beginning of a structural shift that rewrites who controls memory?

The Number That Reframes Everything

¥50.8 billion. That is CXMT's revenue in the first quarter of 2026 alone — roughly $7.5 billion, up 719% from the same quarter a year earlier. Net profit hit ¥24.76 billion. The company lost ¥19.23 billion in 2023. It turned its first annual profit in 2025: ¥7.14 billion. Now it is printing money at a rate that makes the forward price-to-earnings ratio — roughly 5.4× on annualised 2026 earnings — look entirely reasonable next to Samsung and SK Hynix.

This is not normal. This is what happens when a memory supercycle collides with a state-backed capacity build-out that has been a decade in the making. The result is the largest A-share semiconductor IPO in Chinese history, eclipsing SMIC's 2020 listing, and a company that did not exist in any meaningful competitive sense five years ago now sitting within 25,000 wafer starts per month of Micron's total DRAM capacity.

CXMT lists tomorrow. The question is not whether the IPO will pop — with only ~2.2% of shares freely tradable on debut and retail demand at 243× oversubscription, it almost certainly will. The question is whether the capital cycle that delivered CXMT to this moment is about to turn.

What Happened

ChangXin Memory Technologies — now operating as CXMT Corp — priced its STAR Market IPO at ¥8.66 per share (~$1.28), raising ¥57.9 billion ($8.6 billion), with an overallotment option that could push the total to ¥66.6 billion ($9.8 billion). [Tier 1: Reuters, Bloomberg, SCMP]

The offering was 212× oversubscribed by retail investors. Institutional demand was notably softer — a divergence that matters, and we will return to it. [Tier 2: Value Add Pulse, BigGo Finance]

The implied post-listing valuation is ¥579 billion (~$85 billion). That puts CXMT roughly in line with Micron Technology's market capitalisation, despite Micron generating roughly three times the revenue and holding decades of DRAM patents. [Tier 1: Bloomberg, SCMP]

The company plans to use the proceeds for three purposes: ¥7.5 billion for wafer production-line technology upgrades, ¥13 billion for DRAM technology upgrades, and ¥9 billion for forward-looking R&D — including high-bandwidth memory (HBM), the specialised DRAM used in AI accelerators. [Tier 1: CXMT prospectus, via Reuters]

State-owned investors held 36.29% of CXMT before the IPO. Major shareholders include Anhui Investment Group and Phase II of China's National Integrated Circuit Industry Investment Fund — the "Big Fund." Chairman Zhu Yiming, who founded memory-chip designer GigaDevice Semiconductor, has no controlling shareholder and the company has no actual controller. [Tier 1: Reuters, SCMP]

The listing lands in a frenzied context. In the weeks before pricing:

SK Hynix raised $26.5 billion in a Nasdaq IPO on 10 July — the largest US listing by a foreign company in history. Its ADRs opened 14% up, then its Korean shares crashed 15.4% days later in their worst session in nearly two decades. [Tier 1: Reuters, TechCrunch, Bloomberg]
CXMT signed a five-year, $7 billion+ supply agreement with ByteDance and a separate $3 billion+ deal with Tencent in June. Neither company confirmed the figures, but three sources told Reuters the ByteDance deal closed this month. [Tier 1: Reuters]
Apple began testing CXMT DRAM chips for devices sold in China and is lobbying the Trump administration for clearance, despite CXMT's presence on the Pentagon's 1260H list of Chinese military companies. Tim Cook described the memory shortage as a "hundred-year flood." [Tier 1: FT, Bloomberg, MacRumors]
Michael Burry disclosed a short position against Micron at ~$1,052 per share and a broader bearish bet across the semiconductor complex — Nvidia, Applied Materials, the SOXX ETF. [Tier 2: Burry's Substack, via multiple financial outlets]
A crypto startup, Trade.xyz, launched a pre-IPO perpetual futures contract for CXMT on the Hyperliquid decentralised exchange. The contract traded as high as $8.64 — implying a $500 billion+ valuation, roughly 526% above the IPO price — before settling around $6.35. [Tier 2: CNBC, Hyperinsight, Coin Metrics]

What It Actually Means

The memory oligopoly just became a quartet

For two decades, three companies — Samsung, SK Hynix, and Micron — controlled more than 95% of global DRAM production. That concentration was not an accident. It was the product of nine boom-and-bust cycles that wiped out competitors, consolidated capacity, and left survivors with extraordinary pricing power during upcycles.

CXMT at 8% global share — and targeting 17% by 2028 — changes the arithmetic. Not because 8% is enormous, but because the direction of travel is unambiguous. The company went from ~100,000 wafer starts per month to ~265,000 in roughly a year. It plans to hit 350,000 by end-2026 and 500,000 by end-2028. SemiAnalysis projects that would put CXMT within 25,000 WSPM of Micron by year-end. [Tier 2: SemiAnalysis, Tom's Hardware, Citrini Research]

Morgan Stanley estimates China will account for roughly 30% of net new global DRAM wafer capacity through 2028, second only to South Korea. [Tier 2: Morgan Stanley, via Korea Times, MK, BigGo Finance]

This is the capital cycle in action. Memory chips are the most cyclical major semiconductor category. The pattern is well-rehearsed: demand spikes → prices surge → manufacturers race to add capacity → supply overshoots → prices collapse → weaker players exit → concentration increases → the cycle resets. What makes this cycle different is that the capacity being added is not coming from the incumbents alone. It is coming from a state-backed entrant whose investment logic is not purely commercial.

The capital cycle question

Here is the tension at the heart of this story.

On one side: the memory supercycle is real and structural. AI data centres now consume an estimated 70% of all high-end memory chips produced. HBM production consumes 3–4× the wafer capacity of conventional DRAM. Micron's HBM capacity is sold out through 2027. SK Hynix's chairman warned global memory supply will remain roughly 20% below demand through 2030. Gartner projects DRAM prices up 125% for full-year 2026. [Tier 1: TrendForce, Gartner, IDC; Tier 2: SK Hynix chairman via Bloomberg]

On the other side: Samsung, SK Hynix, and Micron have collectively announced more than $500 billion in new capacity investment. SK Hynix alone is spending 1,100 trillion won (~$770 billion) on new fabs in Yongin and Cheongju through 2033. Micron pledged $250 billion in US investment through 2035. CXMT's own expansion — and YMTC's parallel NAND build-out — adds a fourth source of supply that did not exist in previous cycles. [Tier 1: company disclosures; Tier 2: Morgan Stanley, Bloomberg Intelligence]

Bloomberg Intelligence's Shuli Ren argues the global memory shortage likely peaked in Q2 2026, with conditions easing through H2 2026 and into 2027 before potentially tipping into oversupply by 2028. [Tier 2: Bloomberg Intelligence]

Michael Burry is betting on exactly that outcome. His short against Micron — and the broader semiconductor complex — is a wager that the cycle has not been tamed, merely stretched. The forward P/E multiples that make memory stocks look cheap today are, in his view, a trap: they capture peak-cycle earnings that will not persist.

The counterargument, which has the weight of current data behind it, is that this cycle is genuinely different. Long-term supply agreements — Micron has 16 LTAs locking in 20% of DRAM and 33% of NAND shipments, with $22 billion in customer prepayments — create visibility that did not exist in prior cycles. HBM's wafer-intensity means capacity added for AI does not translate neatly into conventional DRAM supply. And the US export controls that block CXMT from acquiring ASML's EUV lithography tools create a technology ceiling that may limit how much of its capacity can compete at the leading edge.

CXMT's IPO, in this reading, is not raising at the top of a cycle. It is raising to fund the catch-up that will determine whether China has a seat at the table when the next cycle begins.

The Quieter Story: Who Is Actually Buying CXMT's Chips

The ByteDance and Tencent deals are the headline numbers — $10 billion+ in combined supply agreements with China's two most important internet platforms. But the more interesting signal is Apple.

Apple does not need CXMT's chips. It has relationships with Samsung, SK Hynix, and Micron that span decades. What Apple needs is a fourth supplier — any fourth supplier — that gives it leverage in price negotiations with the Big Three. Tim Cook's "hundred-year flood" comment is not hyperbole. DRAM contract prices rose 93–98% QoQ in Q1 2026. Apple raised prices on Macs, iPads, and home devices. Memory now accounts for a material share of Apple's bill of materials in a way it has not since the last supercycle.

So Apple is doing something strategically rational and politically explosive: testing CXMT chips for China-market devices while simultaneously lobbying the Trump administration for clearance. The legal situation is nuanced. CXMT is on the Pentagon's 1260H list — which primarily restricts Defence Department contracting, not commercial purchases. It is not on the Commerce Department's Entity List, which would create much harder barriers. Apple does not legally need permission to buy from CXMT. It is seeking political pre-clearance to avoid the backlash that killed its 2022 attempt to source NAND from YMTC. [Tier 1: FT, Bloomberg]

House Foreign Affairs Committee Chairman Brian Mast has already objected: "CXMT and YMTC are Chinese military companies that support the Chinese Communist Party's military modernization and pursuit of AI dominance. If this deal goes through, it would undermine the president's agenda." [Tier 2: Seoul Economic Daily, via Mast statement]

The Apple-CXMT negotiation is a live stress test of whether commercial logic or national security logic prevails in US-China technology policy. The outcome will signal whether the decoupling is hardening or whether carve-outs for specific use cases — China-market devices only, no US supply chain integration — are politically viable.

The Crypto Shadow Market

One of the strangest details in this story is the emergence of a parallel derivatives market for CXMT shares on Hyperliquid, a decentralised exchange. Trade.xyz, a crypto startup, launched a perpetual futures contract under the ticker xyz:CXMT that gives global traders synthetic exposure to CXMT's share price before — and after — the Shanghai listing.

The contract traded as high as $8.64, implying a ~$535 billion valuation — roughly 6.3× the official IPO valuation. It has since settled around $6.35, implying ~$425 billion. [Tier 2: CNBC, Hyperinsight, Coin Metrics]

This is not a real market in any conventional sense. Open interest is modest by Hyperliquid's standards. The contract confers no ownership, no dividends, no voting rights. As one analyst told CNBC: "At this size, I believe the headline valuation is being set by a very modest amount of capital."

But it is a real signal. The premium reflects two things: genuine offshore demand for exposure to a company that most global investors cannot easily access, and the speculative dynamics of a market with no short-side infrastructure to enforce price discovery. When CXMT lists tomorrow, the Hyperliquid contract will re-anchor to the traded price. Gaps like this tend to close abruptly rather than gradually.

The existence of the market is itself a data point about how capital finds its way around regulatory barriers. If you cannot buy CXMT on the STAR Market — which requires a ¥500,000 asset threshold and two years of trading experience for individual investors — you can buy a synthetic derivative on a blockchain. The architecture of global capital markets is becoming stranger and more fragmented, and CXMT's IPO is a case study in that fragmentation.

Stakeholder Landscape

Who wins:

CXMT and its state-backed shareholders. The IPO prices at nearly double the original prospectus target. The company now has ~$8.6 billion to fund capacity expansion and HBM development. The Big Fund and Anhui Investment Group see their stakes marked to market at an $85 billion valuation.
Chinese hyperscalers (ByteDance, Tencent). The supply agreements lock in DRAM at prices reportedly around half the global average — $150 per module versus $300–400 internationally. In a shortage, guaranteed supply at a discount is a structural advantage.
Chinese semiconductor equipment makers. CXMT's expansion — and YMTC's parallel NAND build-out — is accelerating orders for domestically produced fabrication tools, building a Chinese equipment ecosystem that reduces reliance on ASML, Applied Materials, and Tokyo Electron.
Apple (if the deal goes through). A qualified fourth DRAM supplier breaks the Big Three's pricing power. Even if CXMT chips are limited to China-market devices, the mere existence of an alternative supplier changes negotiation dynamics globally.

Who loses:

Samsung, SK Hynix, Micron. A fourth competitor with state backing, growing at 30%+ annual capacity rates, selling at a discount, and locking in China's largest technology buyers on multi-year contracts. The oligopoly's pricing power erodes at the margin — and the margin is where profits live.
Global memory buyers without China access. If CXMT's output is preferentially allocated to Chinese customers — and the ByteDance/Tencent deals suggest it is — the global supply crunch for non-Chinese buyers does not ease. The shortage bifurcates.
Investors long memory at cycle-peak multiples. If Burry and Bloomberg Intelligence are right about the 2028 oversupply inflection, the forward earnings that make today's valuations look reasonable will not materialise.

Who is caught in the middle:

The Trump administration. Caught between national security hawks who want to tighten controls on CXMT and a US company (Apple) arguing that limited commercial engagement is better than ceding the Chinese memory market entirely to a state-backed champion with no Western commercial ties.
ASML and the Dutch government. The US has pressured the Netherlands to restrict advanced lithography sales to Chinese firms since 2019. CXMT's expansion using domestic Chinese equipment — and the projected availability of domestic DUV scanners by late 2026 or early 2027 — suggests the window for export controls to constrain Chinese memory production may be closing.

Cross-Layer Implications

Technology. CXMT is currently producing DDR4 and DDR5 at 19nm — roughly two to three generations behind Samsung and SK Hynix's leading-edge DRAM. The HBM gap is wider. But the company is investing IPO proceeds in HBM development and has built a back-end packaging facility in Shanghai targeting production by end-2026. The technology question is not whether CXMT catches the leading edge — it will not, in this cycle — but whether it can produce "good enough" DRAM for the vast majority of applications that do not require cutting-edge HBM3E or HBM4. The answer, based on Apple's qualification testing, appears to be yes. [Tier 1: Reuters; Tier 2: SemiAnalysis]

Security. The Pentagon's 1260H designation is a political signal, not an operational restriction. But the direction of travel in Washington is toward tighter controls, not looser ones. The House Foreign Affairs Committee has explicitly called for strengthening export controls on CXMT. If the administration denies Apple's request — or if Congress legislates broader restrictions — CXMT's addressable market shrinks to China and non-aligned countries. That does not kill the business. China's domestic DRAM demand is enormous. But it caps the upside that the $85 billion valuation implies.

Commercial. The memory industry's structure is changing from a three-player oligopoly to something more complex. The incumbents are not standing still — their combined $500 billion+ investment plans dwarf CXMT's $8.6 billion raise. But the incumbents are investing at the leading edge (HBM4, sub-10nm DRAM), while CXMT is adding capacity at mature nodes where the volume is. The risk is not that CXMT out-innovates Samsung. It is that CXMT floods the mature-DRAM market with state-subsidised supply, compressing margins for everyone at exactly the moment the incumbents need those margins to fund their leading-edge investments.

Regulatory. The Apple-CXMT negotiation is a regulatory Rorschach test. If the administration approves — even with conditions — it establishes a precedent that 1260H-listed companies can supply US firms under certain circumstances. If it denies, it reinforces the decoupling trajectory and accelerates China's push toward full-stack self-sufficiency. Either outcome has consequences that extend well beyond memory chips.

Geopolitics. CXMT's IPO is the most visible single act yet in China's semiconductor self-sufficiency drive. It is not the only one. YMTC is building a third NAND fab in Wuhan and planning two more. SMEE's domestic immersion DUV scanners are projected to enter volume production around late 2026. The capital cycle in Chinese semiconductors is not a single-company story — it is a national project, and the IPO is the moment it becomes legible to public markets.

What This Means for You

If you buy electronics. Memory prices are not coming down soon. The structural reallocation of wafer capacity toward HBM and enterprise DRAM means consumer-grade memory will remain supply-constrained through at least mid-2027. If you need a laptop or phone, buying sooner rather than later is the rational call. Check the memory specifications carefully — some manufacturers are reducing RAM and storage in mid-range devices while maintaining price points.

If you invest in semiconductors. The CXMT IPO does not change the near-term supply-demand balance — the company's 2026 output is largely committed. What it changes is the medium-term supply picture. Morgan Stanley's 30% estimate for China's share of new DRAM capacity through 2028 is the number to watch. If CXMT executes on its 500,000 WSPM target, the memory market in 2028–2029 looks materially different from today. The Burry short may be early, but it is not irrational.

If you work in technology procurement. The bifurcation of the memory market — Chinese hyperscalers getting preferential access and pricing from CXMT, global buyers competing for the Big Three's remaining conventional DRAM output — is a structural shift, not a temporary dislocation. Multi-supplier strategies that include a China-based option (where politically and legally viable) are becoming a standard part of the procurement toolkit. If your organisation cannot or will not buy from CXMT, you are paying the scarcity premium.

If you follow geopolitics. The Apple-CXMT decision is the next major fork in US-China technology policy. Watch for a decision in Q3 2026. An approval with conditions (China-market devices only, no technology transfer, audit rights) would signal that the administration is willing to carve out commercial exceptions to the decoupling trajectory. A denial would signal the opposite — and would likely accelerate Chinese investment in domestic alternatives across the semiconductor stack.

Uncertainty Ledger

The HBM question. CXMT's ability to produce competitive HBM — the highest-margin, fastest-growing segment of the DRAM market — is unproven. Without HBM, CXMT is a commodity DRAM producer in a market where the incumbents are shifting capacity toward high-margin specialised products. The $85 billion valuation assumes HBM entry. If that assumption fails, the valuation fails with it.
Export control escalation. The US could add CXMT to the Commerce Department's Entity List at any time. That would be a materially different restriction than the 1260H designation — it would block US-origin equipment and technology from reaching CXMT's fabs. The probability of this happening is non-trivial and impossible to price.
The cycle timing. If the memory shortage peaked in Q2 2026, CXMT is listing near the top. If the shortage persists through 2027–2028 as SK Hynix's chairman argues, CXMT is listing into a multi-year tailwind. The honest answer is that nobody knows — including the people running the companies. Memory cycles are forecastable only in retrospect.
The Apple deal. It may not happen. The political opposition is real, the 2022 YMTC precedent is cautionary, and the Trump administration has competing impulses — support US companies versus confront China. A deal failure would not break CXMT (the ByteDance and Tencent agreements provide ample domestic demand) but would cap the narrative upside.
The crypto valuation signal. The Hyperliquid contract's implied $400–500 billion valuation is not a forecast. It is a sentiment indicator in a thin, leveraged market with no short-side infrastructure. Treat it as a curiosity, not a data point.

Bottom Line

CXMT's $8.6 billion IPO is the moment China's memory-chip ambitions become legible to public markets — and the moment the global DRAM oligopoly becomes a quartet. The company is listing into the strongest memory pricing environment in a generation, with revenue up 719% year-on-year and capacity expanding at a rate that will put it within striking distance of Micron by year-end. The capital cycle that made this possible is the same one that poses the existential question: is CXMT raising at the top, or at the beginning of a structural shift that rewrites who controls the world's memory? The answer depends on whether the company can close the HBM technology gap before the cycle turns — and on whether Washington decides that a Chinese memory champion with Apple as a customer is more or less dangerous than one without.

Sources:

Reuters (Tier 1) — CXMT IPO pricing, ByteDance/Tencent supply deals, revenue figures, prospectus details, HBM development timeline
Bloomberg (Tier 1) — IPO pricing, valuation, Apple negotiations, SK Hynix Nasdaq listing, CXMT 1260H designation
Financial Times (Tier 1) — Apple CXMT chip testing, Washington lobbying, 1260H list implications
South China Morning Post (Tier 1) — IPO pricing, Zhu Yiming profile, state-owned shareholder structure
New York Times (Tier 1) — CXMT as China's chip champion, AI race context
Caixin Global (Tier 1) — STAR Market listing details, week-ahead calendar
TechCrunch (Tier 1) — SK Hynix Nasdaq IPO details
CXMT IPO Prospectus (Tier 1) — use of proceeds, shareholder structure, financial disclosures
TrendForce (Tier 2) — DRAM contract price data, Q1 2026 industry revenue, capacity allocation
Gartner (Tier 2) — 2026 DRAM/NAND price forecasts
IDC (Tier 2) — DRAM supply growth projections, device price impact estimates
SemiAnalysis (Tier 2) — CXMT wafer capacity projections, revenue estimates
Morgan Stanley (Tier 2) — China 30% of new DRAM capacity by 2028 estimate
Bloomberg Intelligence / Shuli Ren (Tier 2) — memory shortage peak Q2 2026 analysis
Michael Burry / Substack (Tier 2) — Micron short position disclosure
CNBC (Tier 2) — Hyperliquid CXMT perpetual futures, Trade.xyz
Coin Metrics / Tanay Ved (Tier 2) — Hyperliquid pre-IPO market analysis
BigGo Finance (Tier 3) — institutional/retail demand divergence, A-share memory sector sell-off context
MacRumors / 9to5Mac / TechPowerUp (Tier 3) — Apple CXMT testing and negotiation timeline
Seoul Economic Daily (Tier 3) — Rep. Brian Mast statement on Apple-CXMT