CXMT Overtakes Tencent as AI Memory Demand Drives Chip Rally

CXMT overtook Tencent in market value just 17 days after its Shanghai debut, a reversal now spreading across Google News and global technology coverage. The memory manufacturer ended August 13 worth about $524 billion, while Tencent stood near $511 billion. CXMT reached the top even though its own shares declined that day.

The crossover exposes an unusual divide inside the AI economy. Investors rewarded a company selling memory to data centers while discounting a customer spending heavily to build them. Tencent helped create the demand supporting CXMT, then lost its valuation lead partly because that infrastructure spending weakened its cash flow.

This is not proof that CXMT has surpassed Tencent in revenue, reach, or operating strength. It is a market verdict shaped by tight memory supplies, limited available shares, and expectations that AI infrastructure spending will continue. The ranking says more about today’s appetite for scarce components than the relative quality of the two businesses.

What Changed When CXMT Passed Tencent

CXMT’s rise was a valuation crossover, not an overnight transfer of operating power.

CXMT, formerly known as ChangXin Memory Technologies, makes dynamic random-access memory. DRAM is short-term working memory that lets processors access active data quickly. Servers require large amounts of it to run databases, cloud services, and artificial intelligence workloads.

The company listed on Shanghai’s STAR Market on July 27. Its stock closed at RMB 49 after being sold at RMB 8.66, a first-day increase of 466%. That rally lifted its market capitalization to RMB 3.3 trillion, or roughly $488 billion, according to Shanghai trading data.

The offering raised RMB 57.92 billion, equivalent to about $8.6 billion at the exchange rate cited in the Reuters report. It was the largest mainland Chinese semiconductor offering on record. Only 6.73% of the enlarged share capital was freely tradable when the company listed.

That small float matters. A limited float means relatively few shares can change hands, even when the company’s total valuation covers its entire share base. Heavy demand for a scarce pool of tradable stock can therefore produce unusually large price movements.

CXMT initially became mainland China’s most valuable listed company on its debut. The later comparison with Tencent crossed a broader boundary because Tencent trades in Hong Kong and has long represented China’s largest consumer internet businesses.

On August 13, CXMT ended the session near $524 billion. Tencent’s market capitalization was reported at approximately $511 billion. CXMT shares had slipped 1.2%, but Tencent fell further after its quarterly results shifted attention toward rising AI costs.

That distinction corrects a potentially misleading interpretation of the headline. CXMT did not need another explosive rally to pass Tencent that day. The crossover happened because the two stocks moved differently after investors examined Tencent’s spending and cash generation.

The first-day valuation also varies across reports because share prices changed, currency conversions differed, and some estimates used intraday levels. TechRepublic placed CXMT’s peak market value near $547 billion after the debut. Reuters calculated about $488 billion at the closing price.

Those figures describe different moments rather than a stable appraisal. The durable facts are the RMB 8.66 offering price, RMB 49 first-day close, 466% gain, and restricted initial float.

For readers encountering the story through Google News, the important change is therefore precise. Public markets briefly priced China’s leading domestic DRAM producer above its leading listed internet platform. They did so while AI buyers competed for memory and questioned the near-term returns from infrastructure spending.

Why AI Memory Demand Is Repricing the Supply Chain

The market is rewarding the bottleneck because every AI system needs more than processors alone.

Graphics processors receive most public attention because they perform the parallel calculations behind AI training and inference. Those processors still depend on several types of memory to hold model parameters, intermediate results, prompts, and frequently accessed data.

High-bandwidth memory, or HBM, sits close to AI accelerators and moves data at very high speeds. Conventional server DRAM supports processors and broader system operations. Storage handles information that must persist after a machine powers down.

Demand is rising across this stack. AI models are getting larger, inference workloads are spreading, and agent-based applications can sustain longer computing sessions. Data-center builders consequently need memory capacity alongside accelerators, networking equipment, power systems, and cooling.

CXMT entered public markets during a particularly favorable part of that cycle. Its prospectus said AI demand had fueled the latest memory upswing. The company also warned that slower AI investment or excessive supply additions could weaken the market.

The financial shift has already been sharp. CXMT expected first-half revenue between RMB 110 billion and RMB 120 billion, more than seven times the prior-year level. It forecast net profit between RMB 66 billion and RMB 75 billion after recording a loss in the comparable period.

Separate reporting showed first-quarter revenue growing 719% from a year earlier. CXMT moved from a RMB 2.83 billion loss to a RMB 33 billion net profit during that quarter, according to a review of its IPO financials.

Pricing helped drive that reversal. SemiAnalysis estimates cited in the same report placed CXMT’s average selling prices within 5% to 10% of those charged by Samsung Electronics, SK hynix, and Micron. That gap is much narrower than the traditional image of Chinese memory as a low-cost alternative suggests.

Customers have another reason to diversify. Samsung, SK hynix, and Micron must allocate limited capacity among HBM, server DRAM, mobile memory, personal computers, and consumer electronics. Concentrating resources on higher-value AI products can leave conventional memory markets tight.

TrendForce analyst Ellie Wang told Reuters that memory conditions should remain tight, with price increases expected through the end of 2027. She also said customers seeking more suppliers should create additional opportunities for CXMT.

Results from established producers support the demand thesis. Samsung reported record second-quarter operating profit of 89.5 trillion won, with most of that profit coming from semiconductors. SK hynix recorded quarterly revenue of 60.5 trillion won, according to industry earnings.

Samsung executive Kim Jaejune said demand growth was outpacing production efforts. Samsung and SK hynix together produce about two-thirds of the world’s memory chips, so their supply decisions affect component costs far beyond AI servers.

This is the mechanism behind the chip rally appearing across Google News. AI infrastructure spending does not benefit every participant equally. Suppliers controlling scarce components can raise prices and expand margins, while buyers absorb higher capital requirements before new services produce comparable revenue.

The same dynamic can reach laptops, smartphones, and enterprise servers. Manufacturers compete for production capacity, and higher component costs can alter product configurations or replacement cycles. A data-center memory shortage is not isolated from the rest of the electronics market.

Tencent Is Funding the Supplier That Passed It

The central reversal is that Tencent’s AI spending strengthens CXMT’s order book while placing Tencent’s own cash flow under scrutiny.

Tencent reported second-quarter revenue of RMB 204.8 billion, up 11% from a year earlier. Its domestic games revenue grew 17%, while marketing services revenue increased 22%. Those figures show that its established businesses were still expanding.

Investors focused instead on capital expenditure. Tencent’s spending rose 176% to RMB 52.8 billion as it purchased computing capacity for AI models, agents, and related infrastructure. Free cash flow turned negative at RMB 13.8 billion.

Capital expenditure is money used for long-lived assets such as servers, networking systems, and data-center capacity. It is not automatically a warning sign. The pressure comes from the timing difference between paying for infrastructure and earning durable returns from it.

Tencent must buy capacity before it knows which AI products will attract users, improve advertising, increase game engagement, or support new paid services. Memory suppliers receive revenue earlier because customers need their components before those services can operate.

CXMT reportedly signed a server DRAM agreement with Tencent worth more than $3 billion in June. Reporting also linked CXMT to a five-year ByteDance agreement worth more than $7 billion. These deals indicate that large Chinese platform companies are trying to secure long-term domestic supply.

CXMT had already completed customer validation for advanced DDR5 server memory and started taking orders from Tencent and ByteDance, according to supply-chain reporting. DDR5 is a modern DRAM standard that offers greater bandwidth and capacity than the preceding DDR4 generation.

Server products are becoming more important to CXMT’s business. Their contribution reportedly grew from 8.4% of revenue in 2024 to 26.5% in 2025. That shift connects the company more directly to data-center investment than a business concentrated on personal computers or smartphones would be.

Tencent is therefore both CXMT’s customer and its valuation opponent. Every large order can reinforce expectations for the supplier. The same order can raise questions about Tencent’s spending discipline, utilization, and eventual return on investment.

This does not mean Tencent’s strategy is mistaken. Underinvesting could leave the company without enough computing capacity to compete with Alibaba, ByteDance, Baidu, or international model providers. Capacity shortages could also slow product development and weaken its cloud position.

The market is asking for evidence, however. Tencent needs to show that higher infrastructure spending creates revenue growth, better margins, or defensible user engagement. Until that evidence arrives, component suppliers can capture more immediate economic value.

The contrast resembles other infrastructure cycles. During a rapid buildout, suppliers of constrained equipment often recognize demand before operators prove the profitability of the finished service. Market leadership can shift toward the seller of the scarce input.

That pattern makes the CXMT versus Tencent comparison more useful than a simple ranking. One company represents the physical inputs required by AI. The other represents the applications, platforms, and services expected to justify those inputs.

Readers tracking the story on Google News should watch both sides of that transaction. Rising supplier revenue validates current demand. Improving returns at Tencent would validate the longer chain of assumptions behind it.

Google News Headlines Do Not Settle CXMT’s Technology Gap

A market capitalization above Tencent does not place CXMT ahead of Samsung, SK hynix, or Micron in manufacturing capability.

CXMT held 7.67% of the global DRAM market in 2025, based on sales figures in its prospectus. Other estimates place its 2026 share near 10%. Either measure leaves it well behind the three established global leaders.

Samsung and SK hynix together account for roughly two-thirds of global memory output. Micron adds another major share. This concentration reflects decades of investment, manufacturing experience, customer qualification, and process development.

CXMT’s opportunity is strongest inside China. Domestic cloud companies, device manufacturers, and government-linked buyers want supply that is less exposed to international restrictions. CXMT serves customers across data centers, computers, and smartphones.

Its reported customer portfolio includes Tencent, ByteDance, Alibaba Cloud, Lenovo, Xiaomi, Oppo, Vivo, Transsion, and Honor. That breadth gives the company routes into several large product categories. It also supports China’s policy goal of reducing dependence on imported semiconductors.

However, advanced memory production requires access to specialized equipment. CXMT still lacks the extreme ultraviolet lithography tools used by leading competitors. EUV lithography uses very short-wavelength light to print extremely small chip features with fewer patterning steps.

Manufacturers can produce advanced chips without EUV by using older deep-ultraviolet tools and additional process steps. That approach can increase complexity, cost, and manufacturing risk. Export controls make it harder for Chinese producers to obtain some leading equipment and services.

CXMT also has more to prove in HBM. That category is particularly important for high-end AI accelerators because it supplies the data bandwidth needed to keep processors working efficiently. Leadership in ordinary DRAM does not automatically translate into HBM leadership.

These gaps are why CXMT’s valuation should not be confused with technical parity. A stock can reflect future expectations, national strategic value, or restricted supply. It does not independently measure production yields, energy efficiency, reliability, or product performance.

The initial float adds another layer of uncertainty. With only 6.73% of shares freely tradable at listing, the market price came from a narrow portion of the company’s ownership. That can magnify both enthusiasm and reversals.

Morningstar analyst Jing Jie Yu described the first-day rise as excessive because memory remains cyclical and export controls create a longer-term drag. Hedge fund manager Yuan Yuwei called the stock expensive and questioned whether the optimism was sustainable.

The valuation range among analysts remains unusually wide. One later comparison cited a Nomura target of RMB 116 per share, while Morningstar estimated fair value at RMB 14.90. CXMT traded above three times Morningstar’s estimate when it passed Tencent.

Memory cycles can turn quickly. Manufacturers respond to higher prices by adding capacity, customers reduce purchases after building inventory, and weaker end markets can expose excess supply. CXMT itself lists slower AI investment and competing capacity expansion as risks.

This skepticism does not erase the operational progress. The company has increased its market share, won server orders, improved pricing, and generated substantial profit during the current cycle. It means the valuation discounts further progress that has not yet been independently established.

A careful reading of the Google News narrative therefore separates three claims. CXMT passed Tencent in market value. CXMT has become a serious domestic DRAM supplier. CXMT has not yet shown that it matches the global leaders across advanced manufacturing and HBM.

The Rally Pressures Samsung, SK Hynix, and Micron

CXMT’s scale changes competitive planning even if the company remains behind on leading memory technology.

Samsung, SK hynix, and Micron face two opposing forces. Tight supply supports their prices and margins, but those conditions also give CXMT cash, customers, and investor backing. The longer the shortage lasts, the more resources CXMT can direct toward expansion.

CXMT plans to increase production and has reportedly targeted a 30% DRAM market share by 2030. Reaching that target would require a major change in output, technology, or both. It would also require competitors and regulators to leave enough room for expansion.

The company’s IPO proceeds provide substantial funding for that effort. New factories remain expensive and slow to complete, however. Equipment installation, process tuning, yield improvements, and customer qualification can take years.

Established producers retain important advantages. They already operate large manufacturing networks, hold long customer relationships, and lead in high-value products. Their HBM road maps also connect them to leading accelerator platforms and global cloud companies.

Yet the incumbent position carries its own tradeoff. Prioritizing HBM can constrain supplies of standard server and consumer DRAM. That creates openings for CXMT in products where customers value availability and domestic sourcing.

China’s market gives CXMT a particularly large base. Tencent, ByteDance, Alibaba, and other companies are expanding data-center infrastructure while facing geopolitical uncertainty around imported components. Long-term contracts can reduce supply risk for those buyers and finance capacity for CXMT.

The pressure extends beyond direct sales competition. A viable Chinese supplier can influence negotiations, inventory strategies, and capital plans. Customers gain leverage when they have another qualified option, even if they do not replace an incumbent across every system.

For Micron, the competitive and geopolitical dimensions overlap. CXMT’s valuation reached roughly half of Micron’s market value in one August comparison, despite its smaller global share and technology constraints. That difference shows how strongly investors value access to China’s domestic market.

Samsung and SK hynix face a similar signal. Both are earning record profits from AI demand, but investors are questioning how much capacity they must add and whether future supply will weaken pricing. Chinese competition adds another variable to that calculation.

The likely outcome is not a simple transfer of leadership. Memory customers qualify components carefully because failures can disrupt entire systems. Large buyers often use multiple suppliers to manage risk, pricing, and availability.

CXMT can gain share as an additional supplier without displacing all incumbent products. Its strongest initial role may be supplying Chinese server DRAM while established manufacturers retain leadership in advanced HBM and many international accounts.

That split would still alter the industry. CXMT would capture more of China’s infrastructure spending, reduce reliance on imports, and build manufacturing experience. The incumbents would face a better-funded competitor in one of the world’s largest technology markets.

The story is therefore bigger than one day’s ranking. CXMT’s market debut turned a policy-backed semiconductor project into a publicly priced competitor with billions available for expansion. Its valuation gives management a strategic resource, assuming the share price remains supportive.

Three Signals Will Decide Whether CXMT’s Lead Matters

The next phase depends on operating evidence, not another dramatic valuation headline.

The first signal is CXMT’s ability to deliver qualified server memory at scale. Investors should track server-product revenue, production yields, and repeat orders from Tencent, ByteDance, and other data-center customers. Deliveries that expand without quality problems would strengthen the case that CXMT is becoming a durable supplier.

Customer qualification is more informative than a laboratory benchmark. Data-center operators test memory for performance, error rates, power consumption, thermal behavior, and long-term reliability. Repeat purchases would show that CXMT’s products meet commercial requirements rather than only technical specifications.

The second signal is Tencent’s return on AI spending. Its 176% capital-expenditure increase created the clearest pressure behind the valuation contrast. Future results must show whether that capacity improves advertising, cloud demand, model usage, or engagement across Tencent’s services.

Free cash flow deserves particular attention. A temporary negative period can accompany productive investment, but persistent cash consumption without stronger operating returns would support the market’s preference for suppliers. A recovery would weaken the neat buyer-versus-seller narrative.

The third signal is progress in advanced manufacturing and HBM. CXMT needs credible production milestones, customer validation, or volume shipments that narrow its gap with Samsung, SK hynix, and Micron. Equipment restrictions remain a central constraint, so process efficiency matters alongside nominal product announcements.

Competitor behavior will provide supporting evidence. Faster capacity additions could ease shortages and reduce prices. Continued long-term supply agreements would instead indicate that large customers expect constraints to persist.

Regulatory action also remains relevant. Tighter export controls could limit CXMT’s access to equipment, software, or overseas customers. Looser restrictions would improve its expansion options. Neither outcome changes the present market value, but both affect the earnings needed to support it.

Readers should resist treating every ranking change as a permanent shift. Tencent operates consumer platforms, games, advertising, financial services, and cloud infrastructure. CXMT manufactures a cyclical component whose profitability depends heavily on pricing and utilization.

Market capitalization compresses those different businesses into one number. It is useful for measuring investor expectations, but it cannot resolve whether CXMT’s current profit surge will survive new capacity, weaker demand, or technical obstacles.

The more important conclusion from the Google News headline is structural. AI has made memory strategically visible. A component once discussed mainly by hardware specialists now shapes corporate spending, consumer prices, national policy, and the rankings of the world’s largest companies.

That shift also changes what technology leaders must monitor. AI strategy is no longer only a choice among models or accelerators. It includes memory availability, supplier concentration, infrastructure utilization, and the financial timing of large capital commitments.

Teams following those connections need a reliable way to retain earnings notes, supplier announcements, and policy updates. A searchable AI knowledge base can help connect individual headlines to the evidence behind them.

Watch the next CXMT delivery update beside Tencent’s next cash-flow report. Then compare both with HBM progress and capacity plans from the established memory producers. That combination will show whether the crossover marked a lasting industrial shift or the peak of an unusually hot memory cycle.