137Labs Global

On July 16, 2026, Taiwan Semiconductor Manufacturing Company, the world’s leading pure-play semiconductor foundry, released its second-quarter earnings report, and the reason this set of results attracted such close attention from global capital markets, the semiconductor supply chain, and senior executives across the technology industry was not merely that TSMC is a company of enormous scale and strategic importance, but also that, against the backdrop of continued expansion in artificial intelligence infrastructure spending, intensifying competition in advanced process technologies, and an accelerating restructuring of the global semiconductor supply chain, TSMC has become one of the most important windows through which investors and industry participants can assess the true strength of AI demand, the production bottlenecks affecting advanced chips, and the broader cycle of technology-sector capital expenditure.
Based on the final operating results, TSMC delivered strong revenue, profit, and profitability in the second quarter, while management also provided a relatively positive outlook for growth in the following quarter, which suggests that, despite already elevated market expectations for AI chip demand, the pace of actual order conversion and the utilization of advanced manufacturing capacity remain strong enough to support further expansion in the company’s business scale, while also indicating that major customers, including hyperscale cloud service providers, AI chip designers, and data center operators, have not yet shown clear signs of materially cutting investment or delaying orders.
However, interpreting this report simply as another quarter of earnings growth driven by AI demand would still fail to capture its full significance, because TSMC’s present advantage does not come only from short-term order growth, but rather from a combination of advanced process technology, manufacturing yield, advanced packaging capabilities, customer collaboration, and long-term capital investment, all of which form multiple layers of competitive barriers, and therefore the deeper change reflected in this report is that the global semiconductor value chain is becoming increasingly concentrated among a small number of companies that possess complete technology platforms and large-scale production capabilities, with TSMC positioned at the center of that structural consolidation.
A Set of Results That Far Exceeded Market Expectations
From a financial perspective, TSMC generated approximately NT$1.27 trillion in second-quarter revenue, representing year-on-year growth of about 36%, while net profit reached approximately NT$706.56 billion, an increase of around 77.4% from the same period a year earlier, and at the same time, the company reported a gross margin of 67.7% and an operating margin of more than 60%, figures that not only rank among the highest levels in the company’s history, but also demonstrate that rapid revenue expansion has not been accompanied by any obvious deterioration in profitability caused by the introduction of new process technologies, overseas capacity construction, or rising depreciation expenses.
For a capital-intensive manufacturing company, achieving rapid revenue growth and improving margins at the same time is usually extremely difficult, because large-scale capacity expansion typically brings higher equipment depreciation, research and development spending, labor costs, and yield pressure during the early stages of new production lines, yet TSMC has been able to expand advanced manufacturing capacity while maintaining an exceptionally high gross margin, which indicates not only that demand for leading-edge process technologies remains strong, but also that customers are willing to pay a premium for superior performance, energy efficiency, and reliable supply, thereby allowing the company to offset part of the incremental cost burden through a more favorable product mix and higher utilization rates.
From the perspective of earnings quality, the most important feature of the quarter was not the net profit growth rate itself, but rather the consistency among revenue growth, gross margin expansion, and operating margin improvement, because this combination shows that the company’s core manufacturing business continues to possess substantial operating leverage, meaning that when advanced process production lines operate at high utilization, fixed costs can be distributed across a larger revenue base, unit costs decline, and the rising contribution from higher-end products further lifts average selling prices and overall profitability.
It is also important to note that second-quarter net profit included a certain amount of one-off investment income, which means that the 77.4% year-on-year increase in net profit should not be viewed as a complete representation of the improvement in recurring operating profitability, nor should the same growth rate be mechanically projected into future quarters, but even after excluding non-recurring factors, TSMC’s operating margin, gross margin, and core cash-generating capacity remained strong, indicating that the one-time gain amplified the final profit figure rather than serving as the fundamental reason the company outperformed expectations.
From a capital-market perspective, this kind of earnings structure is generally more persuasive than one in which profit growth is driven mainly by non-recurring gains, because investors are ultimately concerned with whether the core business can continue expanding, whether customer demand is sufficiently stable, whether technological leadership can be converted into pricing power, and whether capital expenditure can generate adequate future returns, and the company’s second-quarter performance provided a largely positive response to all of these questions, thereby reinforcing market confidence in its long-term earnings capacity.
AI Has Become TSMC’s Largest Growth Engine
If smartphones were the primary source of advanced semiconductor demand growth over the past decade, then in the era of rapidly developing generative artificial intelligence, high-performance computing and AI accelerators are increasingly replacing smartphones as the company’s most important sources of growth, and this shift in business emphasis is not merely visible in the revenue mix of a single quarter, but also reflects a fundamental change in the demand structure of the broader advanced semiconductor market.
According to the business mix disclosed by TSMC, high-performance computing now accounts for approximately 66% of revenue, significantly exceeding the contribution from smartphones, and this change suggests that the company’s current growth engine has shifted away from consumer handset replacement cycles and terminal-device upgrades toward the construction of large-scale computing infrastructure by global cloud service providers, internet platforms, AI laboratories, and enterprise customers, a customer group characterized by larger investment budgets, faster technology iteration, higher chip value per unit, and a much deeper dependence on advanced process technologies.
Training and inference for generative AI models require large quantities of high-performance GPUs, specialized AI accelerators, high-bandwidth memory interconnect chips, network switching chips, and server processors, and because these products must deliver greater computing performance within strict limits on power consumption and physical space, they typically rely on the most advanced process technologies as well as increasingly sophisticated packaging solutions, which means that the value AI demand brings to TSMC is not limited to an increase in wafer shipments, but also includes simultaneous growth in demand for higher-priced advanced process nodes, advanced packaging, and related manufacturing services.
From the perspective of the semiconductor supply chain, companies such as Nvidia, AMD, and several major cloud service providers possess significant advantages in chip architecture and software ecosystems, but generally do not own the most advanced semiconductor manufacturing capacity directly, and therefore their ability to launch products on schedule and expand shipment volumes depends heavily on whether TSMC can provide sufficient leading-edge process and packaging capacity, a relationship that makes TSMC not only one of the primary beneficiaries of the AI industry, but also an increasingly important infrastructure provider whose production capabilities influence the pace at which the entire AI hardware market can grow.
More importantly, AI computing demand differs significantly from the demand dynamics that previously characterized the smartphone market, because once smartphone penetration reaches a high level, growth tends to depend on replacement cycles and incremental feature upgrades, whereas AI infrastructure remains in a phase of rapid construction and technological evolution, with larger models requiring continuously increasing amounts of training compute, while the wider adoption of inference services is likely to create a longer-lasting and more distributed form of computing demand, meaning that as long as AI applications continue to expand, related semiconductor demand could gradually extend from centralized model training into cloud inference, enterprise deployment, and edge computing.
At the same time, the fact that AI has become TSMC’s largest growth engine also means that the company’s business mix is becoming more sensitive to the global technology capital expenditure cycle, because if hyperscale cloud companies eventually slow the growth of AI investment, or if AI-related revenue generation fails to justify continuously rising infrastructure spending, demand for advanced chips could be affected, and therefore, although the present growth cycle is supported by a strong industrial rationale, investors still need to monitor the return on capital expenditure among major customers, the pace at which AI applications become commercially viable, and the overall intensity of data center investment.
Advanced Process Technologies Continue to Strengthen TSMC’s Competitive Lead
In addition to sustained AI demand, the rising revenue contribution from advanced process technologies was another of the most important structural highlights in TSMC’s second-quarter results, because for a semiconductor foundry, long-term competitiveness is determined not simply by total manufacturing capacity, but by whether the company can achieve stable mass production, high yields, reasonable costs, and broad customer adoption at the most advanced technology nodes.
From the revenue mix, 3-nanometer technology already accounts for a substantial share, 5-nanometer remains an important source of revenue, and 2-nanometer production has begun to make an initial commercial contribution, while the combined share of 7-nanometer and more advanced processes remains high, indicating that TSMC’s revenue structure has become increasingly concentrated in higher-value nodes, a development that not only supports higher average selling prices, but also deepens the company’s long-term relationships with key customers.
Advanced process technology is particularly important for AI chips because the central bottlenecks in model training and inference often include computing performance, power consumption, thermal management, and data-transfer efficiency, and improvements in process technology allow more transistors to be placed within a similar chip area while enhancing performance and energy efficiency, which means that for data center customers deploying tens of thousands or even hundreds of thousands of accelerators, even a modest improvement in chip efficiency can translate into substantial savings in electricity, cooling, and operating costs.
The gradual commercialization of 2-nanometer production carries especially important strategic significance because it not only marks TSMC’s entry into the next generation of process technology, but could also influence the manufacturing landscape for future flagship smartphone processors, AI accelerators, and server chips, and if the company can continue to lead in yield, capacity, and customer adoption at the 2-nanometer node, then competitors may still find it difficult to replicate TSMC’s scale, manufacturing consistency, and delivery reliability in the near term, even if they are able to achieve broadly comparable technical specifications.
TSMC’s advantage in advanced manufacturing also extends into customer collaboration and the broader design ecosystem, because semiconductor companies often need to work with foundries years in advance to jointly develop process technologies, design rules, and packaging solutions, and once a chip has completed tape-out and entered mass production, switching manufacturing platforms involves significant redesign costs, lengthy validation periods, and supply risks, which means that the customer loyalty created through years of technical cooperation cannot be easily disrupted through price competition alone.
At the same time, competition in advanced process technology is not without risk, because every new node requires enormous investment in research, equipment, and manufacturing infrastructure, while greater process complexity can also lead to longer yield-ramp periods, and if customer products are delayed, order volumes fall below expectations, or implementation costs rise too far, the returns on new production lines may be affected, which means that the successful ramp-up of 2-nanometer production is important not only to TSMC’s technological leadership, but also to its capital efficiency and profit margins over the next several years.
Why Was Gross Margin Still Able to Improve?
TSMC’s gross margin reached 67.7% in the second quarter, and this figure attracted particular attention because the company is simultaneously accelerating investment in advanced process technologies and expanding its global manufacturing footprint, yet has still managed to maintain an exceptionally high level of profitability, suggesting that the positive effects of strong demand, a favorable product mix, and high manufacturing efficiency have so far outweighed the negative impact of new factory costs, rising depreciation, and overseas operating expenses.
First, high utilization rates in advanced manufacturing facilities are a critical foundation for maintaining elevated gross margins, because semiconductor production involves enormous fixed costs, and an advanced fabrication plant must bear expenses related to equipment depreciation, maintenance, energy, personnel, and cleanroom operations regardless of production volume, whereas when customer demand is strong enough to keep production lines operating close to full capacity, those fixed costs can be spread across a greater number of products, significantly reducing unit costs and improving overall profitability.
Second, the continued shift in the product mix toward 3-nanometer, 5-nanometer, and advanced packaging solutions has also raised average selling prices and expanded the company’s profit pool, because these products are typically used in high-performance computing and flagship semiconductor applications, where customers place a greater emphasis on performance, power efficiency, and supply reliability than on minimizing price alone, allowing TSMC to exercise stronger pricing power based on technological leadership and constrained capacity.
Third, the combination of mature and advanced process capacity helps the company balance different industry cycles, because while demand for consumer electronics, automotive chips, and industrial semiconductors may fluctuate at different times, growth in AI and high-performance computing orders can raise utilization rates on advanced production lines, while certain mature-node factories that have already absorbed much of their depreciation can continue to generate stable cash flow, meaning that the company’s overall margin profile is not entirely dependent on a single technology node.
Nevertheless, the second-quarter gross margin of 67.7% may not be sustainable indefinitely, because the early stages of 2-nanometer mass production will likely involve higher manufacturing costs, overseas factories generally face higher labor, construction, and operating expenses than facilities in Taiwan, and as projects in the United States, Japan, and Europe progressively enter production, additional depreciation expenses are likely to become more visible in the income statement, which means that a lower margin outlook in future quarters would not necessarily indicate weaker demand, but could instead reflect the normal cost burden associated with expansion and new-technology introduction.
The key issue to monitor is whether TSMC can use product pricing, cost improvements, capacity utilization, and government subsidies to offset the margin pressure created by overseas manufacturing and the adoption of new process nodes, because if the company can sustain gross margins at relatively high levels over the long term, it would demonstrate that technological leadership has been successfully converted into durable commercial pricing power, whereas if margins decline materially as global expansion accelerates, it could suggest that the internationalization of the manufacturing network, while improving supply-chain security, is reducing capital efficiency.
Management Has Sent a More Positive Signal
Compared with already realized second-quarter results, management’s outlook for the third quarter and the next several years provides a more meaningful indication of the company’s true visibility into customer demand, and TSMC’s revenue guidance and capital expenditure plans were broadly positive, suggesting that the company does not view current AI and advanced-computing demand as a temporary inventory-restocking event, but rather as part of a longer-term infrastructure investment cycle.
The company expects third-quarter revenue in US-dollar terms to continue growing sequentially, which indicates that the strong second-quarter performance is not expected to represent a one-time peak, but may instead extend into the following quarter, while management’s decision to increase capital investment further suggests that existing advanced process and packaging capacity remains insufficient to meet medium- and long-term customer requirements, forcing the company to procure equipment, construct facilities, and prepare additional production capability well in advance.
Semiconductor manufacturing investment cycles are typically long, and the process from site selection and equipment installation to customer qualification and full-scale production often takes several years, meaning that TSMC’s current increase in capital expenditure is effectively a preparation for potential demand two to five years in the future rather than a response to a single quarter’s order fluctuations, and this in turn implies that management must make investment decisions based on customer product road maps, long-term purchasing commitments, and broader industry trends.
From a positive perspective, the increase in capital expenditure indicates that major customers remain confident in future high-performance computing demand, particularly as AI models continue to grow, inference demand accelerates, and custom semiconductor projects become more common, creating the possibility that advanced process and advanced packaging capacity will remain constrained at the same time, and therefore investing early allows TSMC to strengthen customer relationships and reduce the risk that competitors gain opportunities due to supply shortages.
However, higher capital expenditure does not automatically translate into higher future profit, because semiconductor fabrication plants require enormous cash outlays, and once equipment enters production, it generates ongoing depreciation regardless of demand, which means that if future growth falls short of expectations, lower utilization rates could quickly weaken margins and returns on capital, and investors therefore need to assess not only the absolute scale of spending, but also which process technologies are receiving investment, which customers and products the capacity is intended to serve, and when the new facilities are expected to contribute meaningful revenue.
Based on management’s comments, TSMC remains relatively optimistic about the medium- and long-term outlook for AI demand, but the company must still maintain a balance between meeting customer needs and avoiding excessive expansion, because the semiconductor industry has repeatedly experienced periods in which optimistic investment assumptions created later capacity oversupply, while current AI demand, although strong, still faces uncertainty related to commercialization, energy constraints, and policy conditions.
Global Capacity Expansion Has Entered a New Phase
TSMC has continued to expand its manufacturing presence in the United States, Japan, and Europe, and this development indicates that the company’s global strategy is evolving from one centered overwhelmingly on Taiwan toward a more distributed model in which manufacturing capacity is established closer to major customers and strategically important markets, a transition driven not only by commercial demand, but also by supply-chain security, industrial policy, and geopolitical considerations.
For US customers, domestic access to advanced semiconductor manufacturing can reduce geographic concentration risk and improve supply security for government, defense, cloud-computing, and critical-infrastructure applications, which gives both the US government and large technology companies an incentive to support TSMC’s local expansion, while for TSMC itself, building advanced production closer to major customers can deepen commercial relationships and help the company secure more subsidies and long-term order commitments amid growing competition among governments to attract semiconductor investment.
TSMC’s Japanese operations are more closely associated with mature and specialty process demand and are intended to create synergies with the country’s automotive, industrial, and consumer electronics sectors, while the strategic importance of its European projects lies mainly in the automotive and industrial semiconductor supply chain, and although this regionalized production structure can help the company move closer to different customer markets, it also increases management complexity because labor costs, supplier networks, construction efficiency, energy prices, and regulatory requirements differ significantly across regions.
The largest challenge associated with global expansion is cost structure, because Taiwan already has a mature semiconductor cluster, a dense concentration of engineering talent, and a highly developed supplier ecosystem, whereas newly built overseas factories often need to recreate equipment maintenance systems, material supply chains, and workforce training programs, resulting in higher unit production costs, and if these additional expenses cannot be offset through customer premiums, government subsidies, or high utilization rates, they may create sustained pressure on the company’s overall gross margin.
In addition, global expansion may weaken some of the scale efficiencies created by TSMC’s historically concentrated manufacturing system, because a more geographically dispersed production network can improve supply-chain resilience while also increasing coordination, inventory, and cross-regional management costs, which means that the company must establish a new balance between security and efficiency, and the success of this balance will become an important standard for evaluating its internationalization strategy over the coming years.
From a longer-term perspective, TSMC’s overseas strategy is not simply an attempt to replicate its Taiwan facilities in other countries, but rather an effort to create a global network composed of Taiwan-based advanced research and core manufacturing, localized high-end capacity in the United States, mature and specialty production in Japan, and automotive and industrial semiconductor capacity in Europe, and if this system can operate reliably, TSMC’s customer reach and strategic importance will rise further, whereas if overseas projects continue to encounter cost overruns, delays, or labor shortages, the company’s return on invested capital could face significant pressure.
Why Does the Market Remain Optimistic?
The primary reason the market remains optimistic about TSMC is that the company not only benefits directly from rising AI chip demand, but also occupies a position in advanced semiconductor manufacturing that is extremely difficult to replace, and unlike technology companies that depend on a single product cycle, TSMC can serve multiple AI chip designers, cloud service providers, smartphone companies, and high-performance computing customers, allowing it to capture growth from the broader expansion of the industry rather than relying on the success of only one customer or product.
For investors, TSMC increasingly resembles an infrastructure provider to the AI industry, because as different chip designers compete for market share, the company can continue benefiting from total industry demand as long as the winning products still rely on its advanced manufacturing and packaging capabilities, and this position, similar to supplying the essential tools and production infrastructure for an entire sector, gives TSMC a more diversified customer base than that of a single AI chip company.
At the same time, TSMC possesses substantial barriers in technology, customer switching costs, and capital requirements, because new entrants must not only invest enormous sums in fabrication facilities, but also spend years developing process knowledge, improving yields, and earning customer trust, which is why the market generally believes that advanced semiconductor foundry services will remain highly concentrated for the foreseeable future, with TSMC retaining a dominant share.
However, optimistic market sentiment can itself become a source of risk, because once investors already expect the company to maintain rapid growth, a merely “good” earnings report may no longer be enough to support further valuation expansion, and TSMC’s future share-price performance will increasingly depend on whether its results continue to exceed already elevated expectations rather than simply whether revenue and profit continue to grow.
The AI industry also continues to face risks related to energy supply, data center construction cycles, export restrictions, customer concentration, and commercialization returns, and if any of these factors cause major customers to slow capital expenditure, TSMC’s order growth and utilization rates could still be affected on a cyclical basis even if its technological position remains unchanged, which means that any long-term assessment of the company must distinguish between the enduring direction of the industry and the short-term pace of investment.
The market’s optimism toward TSMC is therefore not without justification, but it rests on the assumption that the company can continue maintaining technological leadership, execute the ramp-up of 2-nanometer production, expand advanced packaging capacity, and control the costs of overseas expansion, and any material deviation in one of these areas could force investors to reassess both earnings expectations and valuation.
Conclusion
Taken as a whole, TSMC’s second-quarter earnings report represents far more than another quarter of record revenue and profit, because it also reflects the broader transformation of the advanced semiconductor industry as AI continues to concentrate value in leading-edge process technology, advanced packaging, and large-scale computing infrastructure, while TSMC, through its technological leadership, production capabilities, and customer ecosystem, has become one of the most direct and strategically important beneficiaries of this industry upgrade.
Rapid revenue growth, the rising contribution from high-performance computing, an increasingly advanced process mix, and exceptionally strong gross margins all indicate that the company remains in a period of robust demand and efficient capacity utilization, while management’s decision to expand capital expenditure and broaden the global manufacturing network demonstrates confidence in advanced chip demand over the coming years and a willingness to invest early in order to reinforce long-term competitive advantages.
Nevertheless, the challenges facing TSMC are likely to become more complex than in the past, because the company must not only solve the technical and manufacturing issues associated with 2-nanometer and future process nodes, but also manage cross-border expansion, cost control, customer concentration, geopolitical pressure, and changes in the AI investment cycle, and although these issues may not immediately appear in any single quarterly report, they will ultimately determine the company’s capital returns and earnings stability over the next several years.
From an industry perspective, demand for computing power driven by artificial intelligence remains in an expansionary phase, and as model-training requirements increase, inference applications spread, and enterprise custom-chip projects become more common, the strategic importance of advanced process technologies and advanced packaging is likely to continue rising, meaning that as long as global AI infrastructure investment does not undergo a major reversal, TSMC will remain one of the most important manufacturing platforms in the entire semiconductor value chain.
For that reason, the most accurate interpretation of this earnings report is not simply that “TSMC delivered strong results” or that “AI demand remains robust,” but rather that the company is evolving from a traditional semiconductor foundry into an increasingly indispensable manufacturing foundation for the global digital economy and AI infrastructure, and its future value will depend not only on whether the AI industry continues to expand, but also on whether TSMC can preserve the execution capability and profitability it has built over many years while navigating high capital intensity, global operations, and rapid technological change.

Avez-vous été submergé par les informations concernant Changxin Technology aujourd’hui ? Vers le 15 juillet, la nouvelle selon laquelle la plateforme Hyperliquid (construite sur le protocole Trade.xyz HIP-3) était sur le point de lancer des contrats perpétuels CXMT a rapidement dominé les discussions des groupes d’investissement et les réseaux sociaux.
En tant que plus grand fabricant chinois de puces mémoire DRAM et le quatrième plus grand au monde, Changxin Technology a fixé son prix d’offre à 8,66 RMB par action le 14 juillet. Les souscriptions en ligne et hors ligne commenceront le 16 juillet et la société devrait être cotée sur le marché STAR de la Bourse de Shanghai vers le 27 juillet (code boursier : 688825.SH). Il ne s’agit pas seulement de l’événement d’introduction en bourse le plus important du marché des capitaux chinois en 2026, mais aussi du deuxième plus important de l’histoire du marché STAR et du plus grand lancement de nouvelles actions dans les actions A cette année, avec une levée de fonds prévue de 295 milliards RMB, ce qui en fait l’une des plus grandes introductions en bourse d’Asie cette année.

Le 14 juillet (heure locale), le Bureau of Labor Statistics (BLS) des États-Unis a publié le rapport sur l’indice des prix à la consommation (IPC) pour juin 2026. En tant qu’un des indicateurs macroéconomiques les plus suivis dans les marchés financiers mondiaux chaque mois, l’IPC reflète non seulement l’évolution des prix à la consommation aux États-Unis, mais sert aussi de référence clé pour évaluer l’orientation future de la politique monétaire de la Réserve fédérale. Par conséquent, investisseurs, économistes, décideurs politiques et acteurs des marchés avaient tous suivi de près la publication. Les dernières données ont montré que l’inflation globale s’est nettement apaisée par rapport aux mois précédents, avec un IPC global (headline) et un IPC hors éléments volatils (core) en dessous des attentes du marché. Les chiffres ont immédiatement stimulé le sentiment des marchés : les principaux indices boursiers américains ont progressé, les rendements des bons du Trésor ont reculé et le dollar américain s’est affaibli, les investisseurs ayant revu à la baisse leurs anticipations de nouveaux resserrements monétaires de la Réserve fédérale. Néanmoins, malgré ce regain d’optimisme, une question plus fondamentale s’est imposée : cette amélioration signifie-t-elle que l’inflation aux États-Unis est enfin maîtrisée, ou s’agit-il simplement d’une baisse temporaire liée à des facteurs de court terme ?

South Korea's stock market is experiencing a dramatic roller-coaster ride in 2026. The KOSPI index surged more than 85% in the first half of the year, hitting a record high of 9,385 points in June. However, in mid-July, due to massive leverage liquidations and foreign capital outflows, it quickly fell into a technical bear market.
From its peak, the index has declined approximately 27% (closing around 6,857 points as of July 14), while SK Hynix (000660.KS) has dropped over 30% from its recent high (with a single-day maximum decline of 15.4%). Margin debt once hit a record 38.6 trillion KRW.
This report systematically analyzes the turning point and opportunities in the Korean market through four dimensions: recent performance and historical data, trading systems and leverage mechanisms, high-quality non-semiconductor targets, and investment strategies with risk warnings.
Recent Performance and Macro Overview
In 2026, South Korea's stock market has seen breathtaking volatility. At the beginning of the year, the KOSPI hovered around 4,300 points. Driven by the global AI wave and semiconductor recovery, the index surged over 85%, reaching a record 9,385.59 points in June. Total market capitalization surpassed the UK, ranking it as the world's eighth-largest stock market.
However, the market reversed sharply in July. On July 13, the KOSPI plunged 8.95% in a single day, officially entering a technical bear market with a pullback of over 27% from its peak. SK Hynix and other heavyweight stocks led the decline, triggering widespread panic selling across the market.
Historical Data and Market Scale
The KOSPI index was first calculated on January 4, 1980 (base value 100 points). As of April 2026, South Korea's total stock market capitalization was approximately $4.04 trillion. The market is clearly dominated by chaebols, with the five major groups — Samsung, SK, Hyundai, LG, and CJ — accounting for over 50% of total market cap.
Trading System and Settlement Mechanism
South Korea's stock market trades from 09:00 to 15:30 (KST), Monday to Friday. Retail investors generally enjoy very low or zero commissions, with the main cost being securities transaction tax (0.08%-0.23%).
The current settlement cycle is T+2, with funds and securities formally settled by 16:00 on the second business day after the trade. The government is actively pushing to shorten it to T+1, with a target implementation in 2027 to improve efficiency and reduce settlement risk.
Leverage Frenzy and Deleveraging Crisis
The 2026 rally was largely driven by retail investors using high leverage. By late June, margin debt reached a record 38.6 trillion KRW (about $260 billion), doubling from early 2025. With a population of about 52 million, the average leverage per person exceeds 1 million KRW. Notably, people over 60 account for one-third of margin debt, and new accounts opened by minors have surged nearly tenfold.
Leveraged ETFs are enormous, totaling about $469 billion and accounting for 1.5% of free-float market cap. The two-times leveraged ETF tracking SK Hynix alone has assets of $130 billion, significantly amplifying volatility.
In July, SK Hynix's decline triggered a chain of forced liquidations. Broker margin maintenance ratios range from 40% to 100%. When prices fell more than 30%, liquidations became widespread. Single-day forced liquidation funds exceeded 4 trillion KRW. The Korean Finance Minister convened an emergency meeting, pointing out that high semiconductor concentration and excessive leverage are amplifying systemic risk.
Growth Opportunities in Other Areas of the Korean Stock Market
In addition to Samsung Electronics and SK Hynix, South Korea has globally competitive leaders in other industries. The following provides a comprehensive review from dimensions such as revenue growth, market cap share, core drivers, valuation, policy support, and specific strengths and advantages (data as of mid-July 2026):
Hyundai Motor Group (Hyundai Motor 005380.KS, Kia 000270.KS)
2025 group combined revenue approximately 300 trillion KRW, market cap share ~8-10%. EV sales achieved explosive growth in 2025-2026 (annual growth >30%).
Strengths and Advantages: World's fifth-largest automaker with strong competitiveness in EVs and hybrids, high export ratio, and resilience to geopolitical risks; clear vertical integration advantages (battery + vehicle + smart platform) and close cooperation with Samsung; successful brand rejuvenation with continuous overseas market share gains; reasonable valuation compared to semiconductors, stable cash flow, and long-term deterministic growth in the global automotive electrification wave; strong resilience during corrections, making it a core defensive + growth pick for diversifying away from heavyweight risk.
LG Energy Solution (373220.KS)
2025 revenue approximately 23.7 trillion KRW, 2026 target year-on-year growth 10-20%, market cap share ~4-5%.
Strengths and Advantages: World's second-largest power battery manufacturer with leading 46-series cylindrical battery and LFP technology, rapidly increasing ESS business share; dual drivers of AI data center energy storage + EV demand, rapid North American capacity expansion, and extremely high visibility of long-term orders; fundamentals highly independent from the semiconductor cycle, with clear revenue and profit recovery in 2025-2026; significant technological barriers and scale advantages, plus policy incentives from North American localization; strong growth certainty, making it one of the most explosive targets in the current new energy track.
Industrial Automation & Robotics Sector (HD Hyundai Electric 267260.KS, etc.)
Sector overall revenue CAGR above 15%, HD Hyundai Electric revenue stable at several trillion KRW level, market cap share ~1-2%.
Strengths and Advantages: Core beneficiary of South Korea's manufacturing "unmanned" and smart factory transformation, with explosive demand for AI-powered collaborative robots; complete industry chain and high order stability; strong government support under the "Smart Manufacturing 2030" policy and high export potential; reasonable valuation and strong anti-cyclical ability, making it the purest manufacturing growth sector outside of heavyweights; provides high-certainty medium-to-long-term growth opportunities amid global supply chain restructuring.
NAVER (035420.KS) and Kakao (035720.KS)
2025 combined revenue approximately 18 trillion KRW, market cap share ~6-8%.
Strengths and Advantages: Deep local internet ecosystem moats — NAVER dominates search, e-commerce, and content, while Kakao Talk has extremely high user stickiness; rapid growth in AI cloud services, digital content, and fintech businesses with abundant cash flow; maintains dominant local position against global giants with steady overseas expansion; relatively reasonable valuation with both growth and defensive attributes, serving as the anchor of South Korea's consumer internet sector.
Amorepacific (090430.KS)
2025 revenue approximately 6-7 trillion KRW, explosive export growth (North America and Europe sales doubled), market cap share ~2%.
Strengths and Advantages: Global leader in K-Beauty with rich brand matrix, high R&D investment, and fast new product iteration; successful e-commerce and overseas channel transformation with continuously rising export ratio and strong resistance to domestic consumption fluctuations; outstanding brand premium, highest elasticity during consumption recovery cycles, and stable cash dividends — a classic defensive consumer leader.
KB Financial Group (105560.KS)
2025 revenue approximately 15 trillion KRW, market cap share ~3-4%.
Strengths and Advantages: Leading financial holding group with stable banking business and strong synergies in insurance and asset management; has launched the "Corporate Value-up Program" to enhance dividends, buybacks, and shareholder returns with significant policy dividends; stable net interest margin, low non-performing loan ratio, and historically low valuation; high dividend + low valuation combination provides stable returns in a high-interest-rate environment — a classic defensive + value re-rating target.
Overall Summary: These companies collectively account for approximately 25-35% of market cap. Most showed dual revenue and profit growth in 2025-2026 (especially explosive export growth at LG Energy Solution, Hyundai Motor Group, and Amorepacific). Their growth is independent of the semiconductor cycle, with more attractive valuations, higher dividend yields, and stronger risk resistance — making them suitable for diversified allocation in the current high-volatility environment.
Summary and Outlook
The Korean stock market is in a transition period from leverage-driven growth to fundamental re-rating. In the short term, deleveraging pressure dominates — investors are advised to stay cautious or lightly position in value stocks. In the medium term, AI/HBM, EV transformation, and the Value-up program remain the main themes, with potential for KOSPI to gradually recover to the 8,000-9,000 range. For long-term investors, the current correction may present a good window to allocate quality non-heavyweight assets.
Data Sources: KRX, Trading Economics, Bloomberg, Ministry of Economy and Finance of Korea, etc. (as of July 14, 2026).
Disclaimer: This article is for informational purposes only and does not constitute any investment advice. The market is highly volatile — please conduct your own research and bear the risks independently.

Introduction
Over the past two years, market discussions about artificial intelligence have focused primarily on whether model capabilities can continue to improve, whether generative AI can become a general-purpose technology comparable to the internet and smartphones, and how large language models may transform industries such as search, software, advertising, and enterprise services; however, as model performance has advanced, user adoption has accelerated, and global computing infrastructure has entered an intensive construction phase, investor attention has shifted increasingly toward commercial returns, with the central question now being whether the enormous amounts of capital that major technology companies are investing in AI can ultimately be converted into stable revenue, profits, and free cash flow.
In its report, Why AI Companies May Invest More than $500 Billion in 2026, Goldman Sachs noted that the consensus forecast for 2026 capital expenditure among leading AI hyperscalers had been raised from $465 billion to $527 billion, while broader estimates suggest that global AI infrastructure investment could continue to expand rapidly over the coming years. As capital moves beyond chips and servers into data centers, electricity generation, cooling systems, and network infrastructure, AI is no longer simply a cycle of software and semiconductor innovation, but is increasingly becoming a global restructuring of capital across energy, industry, finance, and infrastructure.
I. Why AI Capital Expenditure Continues to Rise
The reason major technology companies continue to increase AI investment despite pressure on free cash flow, concerns about valuation, and periods of weak share-price performance is that they do not view AI as an ordinary product investment, but as a strategic commitment that may determine leadership in the next decade of computing platforms, cloud services, and software ecosystems, which means their greatest concern is not whether they spend several billion dollars too much in a particular year, but whether they lose their competitive position as the next generation of technological infrastructure takes shape.
Microsoft, Amazon, Alphabet, and Meta are not facing a conventional cost-benefit decision, but a strategic contest in which any company that slows investment first risks seeing cloud customers move to competitors with greater computing capacity, developers migrate toward rival ecosystems, and its own capabilities in model training, inference, and product development gradually fall behind.
A substantial portion of current AI capital expenditure therefore has a defensive character. Even though technology companies have not yet proved that every dollar invested in AI will generate an attractive return, they generally believe that the long-term cost of failing to invest may be greater. Once all leading companies reach a similar conclusion, the industry enters a competitive structure resembling a prisoner’s dilemma, in which every participant would prefer its rivals to reduce spending, yet no market leader is willing to withdraw first.
At the same time, the market’s understanding of computing demand is also changing, because AI infrastructure investment initially focused mainly on training large models, whereas the emergence of AI agents is now making inference demand an increasingly important source of growth. A conventional chatbot may require only one or several model calls to answer a user’s question, but an AI agent performing a genuinely complex task may need to break down objectives, search for information, compare alternatives, invoke tools, revise its plan, and repeatedly verify its output, meaning that its consumption of tokens and computing resources may be several or even dozens of times greater than that of a standard question-and-answer interaction.
Goldman Sachs estimates that, driven by the adoption of AI agents by both consumers and enterprises, global token usage could increase twenty-fourfold between 2026 and 2030, reaching 120 quadrillion tokens per month. The logic behind this forecast is that AI may no longer remain a tool that users open occasionally, but could gradually become a persistent execution system embedded in office software, customer service, advertising, financial analysis, e-commerce, software development, and industrial processes.
If this transition occurs, the main driver of future computing demand will shift from a limited number of companies periodically training large models to hundreds of millions of users and enterprises continuously running intelligent agents, while inference computing will increasingly resemble the recurring consumption patterns of cloud computing, electricity, and telecommunications traffic. As a result, even if training efficiency improves or individual models are trained less frequently, total demand for computing resources across the economy may still grow rapidly.
Another Goldman Sachs study suggests that improvements in chip performance, model compression, and data-center architecture could reduce the unit cost of AI inference per token by 60% to 70% annually, but lower unit costs do not necessarily imply lower total industry spending, because technological history repeatedly shows that when a resource becomes cheaper, its range of uses and total consumption can increase rather than decline, and AI may follow the same pattern.
As inference costs continue to fall, applications that were previously uneconomic, including automated customer service, real-time video generation, personalized education, software development agents, and enterprise process automation, may become suitable for large-scale deployment. The key variable determining the future scale of AI capital expenditure is therefore not the price of an individual token, but whether growth in usage can continue to exceed the decline in unit cost.
II. Why the $527 Billion and $765 Billion Estimates Reflect Different Measurement Frameworks
To understand Goldman Sachs’ assessment of AI investment, it is necessary to distinguish between the statistical scope of the $527 billion and $765 billion figures. The former refers mainly to the 2026 consensus capital expenditure forecast for leading AI cloud companies such as Microsoft, Amazon, Alphabet, and Meta, including spending on servers, data centers, networking equipment, and other capital projects; the latter comes from Goldman Sachs’ broader AI infrastructure model, which includes not only direct investment by technology companies, but also the data-center buildings, electricity supply, cooling systems, and other infrastructure required to support those computing systems.
The $527 billion figure is therefore closer to the annual capital budgets of leading technology companies, while the $765 billion estimate more closely represents the annual construction cost of the entire AI infrastructure ecosystem, and the projected $7.6 trillion reflects potential cumulative investment between 2026 and 2031.
Although chips receive the greatest attention within AI infrastructure, they are only the starting point of the system, because a large-scale AI computing cluster also requires server racks, high-speed networking, optical components, liquid-cooling systems, uninterruptible power supplies, transformers, transmission lines, backup generation, and the data-center buildings needed to house and operate all of these assets. Every new deployment of AI accelerators therefore tends to require system-wide investment that may be several times greater than the value of the chips themselves.
Some of the baseline assumptions used in Goldman Sachs’ model include approximately 3,000 watts of package power for an AI accelerator, a data-center power usage effectiveness ratio of around 1.2, data-center construction costs of roughly $15 million per megawatt, and incremental power-infrastructure costs of approximately $2,500 per kilowatt. These parameters indicate that competition in AI is no longer limited to chip design and model development, but is increasingly becoming a broader contest involving land, electricity, financing, supply chains, engineering capacity, and long-term operating expertise.
III. Why Semiconductor Companies Are Already Profitable While the Broader AI Industry Has Yet to Prove Its Returns
So far, the clearest and most concentrated economic value in the AI industry has emerged among semiconductor companies and related equipment suppliers, because major technology companies must purchase chips, build data centers, and complete networking and power infrastructure before AI businesses generate stable revenue, while chip suppliers can recognize revenue and profit shortly after delivery. By contrast, cloud providers and model developers that purchase these chips must recover their investment over many years through cloud services, software subscriptions, improvements in advertising efficiency, or enterprise AI products.
The result is that profits across the AI value chain are currently concentrated among upstream chip and equipment suppliers, while cloud platforms, model companies, and enterprise customers remain in the investment and commercialization-validation stage, meaning that the industry as a whole has not yet formed a broad and stable cash-flow cycle.
Goldman Sachs has noted that although semiconductor companies are generating record levels of revenue and profit, many other participants in the AI ecosystem have not yet achieved returns commensurate with their investment, and a structure in which upstream suppliers make money first while downstream customers continue to spend cannot be sustained indefinitely. A healthy value chain must ultimately be supported by end customers that use AI to increase revenue, reduce costs, or improve efficiency, because only when profits gradually spread from chip manufacturers to cloud platforms, software companies, and end applications will customers be able to keep increasing purchases and drive the industry into a stable positive cycle.
The consumer market has delivered extremely rapid AI adoption, but user growth cannot be equated directly with commercial value. Research cited by Goldman Sachs suggests that generative AI reached an adoption rate of approximately 53% within three years of the launch of the first widely available product, significantly faster than the early adoption of personal computers or the internet, yet many users rely on free products, and even when some consumers pay subscription fees, that revenue may not be sufficient to cover model training, inference, electricity, data centers, and research and development.
The customers most capable of supporting trillions of dollars in AI infrastructure investment are therefore likely to be enterprises, because businesses have greater purchasing power and operate large-scale processes in customer service, sales, research and development, finance, and supply chains, although enterprise AI deployment is far more difficult than consumer use of a chatbot. Many companies initially believed that purchasing the most advanced model would automatically generate productivity gains, but in reality their internal data is often fragmented across different systems, with inconsistent formats, unclear permissions, and uneven quality. If inventory, membership, order, and recommendation data remain disconnected, even a highly capable model may struggle to produce stable and reliable business outcomes.
The key constraint on enterprise AI adoption is therefore no longer model capability alone, but whether companies can complete data governance, model orchestration, and business-process redesign. Enterprises are likely to use different models according to task complexity, cost, data security, and risk, while also establishing permission controls, human review, and output-tracing mechanisms, which means enterprise AI is not simply a matter of purchasing a software license, but a long-term project involving system redesign, compliance review, and organizational change.
Goldman Sachs estimates that only around 12% of knowledge workers may use AI agents by 2030, with the figure potentially rising to 37% by 2040, suggesting that infrastructure investment may advance significantly faster than the realization of enterprise commercialization returns.
IV. Is AI Investment Becoming a New Bubble?
Comparing the current AI boom with the internet bubble of the late 1990s has become one of the most common market frameworks, because both periods have involved rapid increases in capital expenditure, highly optimistic assumptions about future demand, companies building infrastructure in advance out of fear of missing a technological revolution, and valuations that depend heavily on revenue and profit expected many years into the future. Nevertheless, concluding from these similarities alone that AI must repeat the dot-com collapse would also be overly simplistic.
Today, the main AI investors are global technology companies with substantial operating cash flow, mature business models, and strong balance sheets, whereas many telecommunications and internet companies during the dot-com era relied heavily on leverage and had not yet established stable revenue or profits. The current AI investment cycle may therefore be less likely to end in a broad systemic collapse, although that does not eliminate the possibility of capital misallocation and structural overcapacity.
The first risk is that investment may move faster than commercialization. Data centers, chips, and power facilities can be built within several years, whereas enterprise AI revenue, process redesign, and organizational transformation may require much longer to mature. If supply is created well in advance of effective paying demand, returns on capital may remain under pressure for an extended period.
The second risk concerns the uncertain economic life of AI chips. Goldman Sachs has identified accelerator lifespan as one of the most important variables determining cumulative AI investment, because AI accelerators are generally expected to operate for four to six years, while newer generations may replace them rapidly with higher performance and lower unit costs. By comparison, data-center buildings may be depreciated over approximately twenty years, while power infrastructure can remain in use for twenty-five years or longer. If chips require frequent replacement, technology companies will face not only recurring investment requirements, but also greater depreciation pressure and a higher risk of equipment obsolescence.
The third risk lies in electricity and engineering constraints, because even after companies purchase large volumes of chips, those assets cannot operate efficiently if data centers fail to obtain power connections on schedule, or if transformers, transmission lines, generation equipment, and cooling systems are not built at the same pace. The main bottleneck in AI development may therefore shift gradually from chip shortages to electricity, land, regulatory approvals, and construction capacity.
Beyond industrial investment risk, capital markets also face the danger of excessive valuation expansion. Goldman Sachs research showed that its basket of AI infrastructure stocks at one point generated an average year-to-date return of approximately 44%, while consensus earnings-per-share forecasts for those companies over the following two years increased by only around 9%, indicating that a substantial portion of share-price appreciation came from multiple expansion rather than corresponding upgrades to profit forecasts.
For valuations to continue rising, companies would need to deliver earnings well above expectations, receive even higher valuation multiples, or demonstrate stronger market power and pricing ability. If capital-expenditure growth slows or downstream customers reduce orders, companies whose share prices have risen much faster than their earnings expectations may face more severe valuation compression.
V. AI Investing Is Moving from Thematic Trading to Fundamental Differentiation
Goldman Sachs found that the average share-price correlation among major publicly listed AI hyperscalers fell from roughly 80% to 20%, indicating that the market no longer treats all AI companies as part of a single trading theme, but is increasingly distinguishing among them according to capital expenditure, financing structure, revenue realization, and cash-flow quality.
During the early phase of the AI rally, investors were mainly concerned with whether a company belonged to the AI value chain, but as valuations and investment levels rose, the market began asking more demanding questions, including whether capital expenditure had already produced revenue, whether a company depended on debt financing, whether customer demand was stable, and whether its business model had a clear path to profit realization. AI investing is therefore moving from a broad thematic trade toward more rigorous fundamental selection.
Future market analysis is likely to focus heavily on the relationship between capital expenditure and AI revenue, because rising investment is not necessarily a problem if revenue and profit expand at the same or a faster rate. Investors will also monitor free cash flow, since a company can continue to report profit growth while free cash flow declines if capital expenditure rises more quickly. Financing structure will become increasingly important as well, because funding investment through operating cash flow creates a very different risk profile from relying on debt to build data centers.
Computing utilization will also determine whether data-center investment can generate adequate returns, because the number of installed servers does not itself represent effective demand. Only when computing assets maintain consistently high utilization can companies cover depreciation, electricity, and maintenance costs, while businesses with cloud platforms, developer ecosystems, proprietary data, and stable enterprise relationships are more likely to convert infrastructure advantages into recurring revenue and customer lock-in.
VI. The Next Winners May Extend Beyond Semiconductor Companies
Goldman Sachs believes that after semiconductor companies captured the first major share of AI profits, the next stage of value creation may gradually shift toward hyperscale cloud providers, AI platforms, and companies that benefit from productivity gains, because the market already understands the capital-expenditure burden facing technology giants, but may underestimate the profit potential associated with rising AI cloud revenue and declining unit costs.
As inference efficiency improves and usage expands, cloud platforms may be able to achieve both revenue growth and lower unit costs, gradually developing economies of scale similar to those of traditional cloud computing. This suggests that platform companies with substantial computing resources, established customer bases, and broad software ecosystems may generate more stable returns as AI commercialization matures.
At the same time, the orchestration layer connecting enterprise data, business processes, and different models may become a new center of value creation. Platforms capable of providing data integration, model routing, cost control, permission management, and compliance auditing could assume an infrastructure role comparable to databases and middleware in the cloud-computing era, while earning stable revenue through high switching costs.
The economic value generated by AI may also accrue not only to AI suppliers, but to companies that use AI to improve their core operations. When an advertising company uses AI to improve conversion rates, a logistics company applies AI to reduce transportation costs, or a software company shortens development cycles through AI-assisted coding, the resulting value may appear directly in the income statements of those users. The next phase of AI investing should therefore focus not only on companies selling AI, but also on businesses capable of using AI to redistribute industry profits, expand market share, or improve cost structures.
AI infrastructure investment will also create opportunities for many industries outside traditional technology. Goldman Sachs estimates that major technology companies could spend a cumulative $5.3 trillion on related capital expenditure between 2025 and 2030, up from an earlier forecast of $4.5 trillion, and a construction program of this scale cannot be financed entirely from technology companies’ internal cash flow.
Data-center developers, power companies, equipment suppliers, infrastructure funds, real-estate capital, and credit markets will all become important participants. As of September 2025, global infrastructure funds managed more than $1.7 trillion in assets and held approximately $400 billion in unallocated capital, while related assets under management could exceed $3 trillion by 2030, suggesting that AI may become an important growth driver for private markets, infrastructure finance, and corporate bond issuance over the coming years.
VII. How to Determine Whether AI Capital Expenditure Will Ultimately Succeed
Assessing whether the current AI capital-expenditure cycle is sustainable requires more than examining chip sales, server orders, or data-center construction, because the analysis must consider demand, revenue, unit economics, returns on capital, and the distribution of profits across the value chain.
Investors must first determine whether AI user growth, enterprise adoption, and token usage continue to expand, while also assessing whether cloud platforms, model companies, and software providers can convert usage into paid revenue, and whether the income generated by each inference task or AI application is sufficient to cover chip depreciation, electricity, networking, maintenance, and research and development. Only when genuine demand, commercial revenue, and viable unit economics exist simultaneously can infrastructure investment receive durable support.
Investors must also evaluate whether the incremental operating profit generated by AI exceeds the cost of the capital required to build the supporting infrastructure, because revenue growth does not necessarily create value if each dollar of AI revenue requires more than one dollar of incremental investment.
Ultimately, the maturity of the AI industry will depend on whether the entire value chain forms a stable commercial cycle rather than on whether chip suppliers alone continue to earn high profits. If model companies, cloud platforms, and enterprise customers can use AI to generate sustainable revenue and improve cash flow, current capital expenditure may be converted into long-term economic value; if downstream companies remain dependent on continuous investment without producing adequate returns, the sustainability of the entire cycle will come into question.
Conclusion
Goldman Sachs’ forecast of more than $500 billion in AI capital expenditure reveals not only how much technology companies may continue to spend, but also that the global economy is entering a new investment cycle driven jointly by computing capacity, data centers, and power infrastructure. Historical experience suggests that major technological revolutions often produce genuine productivity gains and temporary capital misallocation at the same time, as railways, the internet, and fiber-optic networks all changed the world while imposing heavy losses on investors whose projects lacked viable commercial returns. The central uncertainty surrounding AI is therefore no longer whether the technology will continue to advance, but which companies will be able to convert technological advantages into durable revenue, profits, and cash flow.
From this perspective, $500 billion may represent both the beginning of a new cycle and the first dividing line in the market’s assessment of AI’s commercial viability, while the ultimate success or failure of this investment wave will depend not on the absolute size of capital expenditure, but on whether those investments can generate long-term economic value sufficient to cover their cost of capital.

It has been three days since South Korean memory giant SK Hynix made its U.S. listing debut. This article reviews the event through four key angles: the listing details and business strengths of SK Hynix, the support from surging AI capex for the memory sector, Goldman Sachs’ breakdown of the AI value chain, and SK Hynix’s strategic value with forward-looking insights.
SK Hynix priced its ADRs at $149. Trading began under SKHYV and later switched to SKHY. The company raised a record $26.5 billion in the largest-ever U.S. IPO by a foreign company. Its ADRs surged about 13% on the first day, pushing market cap above $1.2 trillion. Proceeds are mainly for expanding wafer fabs in Korea and acquiring advanced EUV equipment to accelerate HBM capacity buildout. As a key supplier to NVIDIA and other high-end AI accelerators, its products already hold a critical position in global data center supply chains.
SK Hynix’s Core Strengths and Industry Position
SK Hynix has long specialized in DRAM and NAND, building deep expertise in HBM technology. Its HBM3E products are now in volume production, and the HBM4 roadmap is advancing smoothly. Since 2025, explosive demand for high-bandwidth memory in AI servers has driven strong order growth, with multiple cloud providers locking in future capacity early.
The global memory market is dominated by an oligopoly of SK Hynix, Samsung, and Micron. HBM shows even higher concentration. The stringent bandwidth and capacity requirements of AI training and inference have made HBM an essential component for data centers. SK Hynix maintains a leading edge through technological advantages and stable customer relationships.
AI Capex Surge Supporting the Memory Sector
In 2026, global AI infrastructure construction has entered a high-density phase. Major hyperscalers including Amazon, Microsoft, Alphabet, and Meta plan combined capital expenditures of approximately $725 billion, a sharp year-over-year increase. Most of this spending targets data centers, server clusters, and high-end chips, creating steady demand for memory products.
SK Hynix’s HBM is a major beneficiary of these expenditures. Cloud providers favor long-term agreements with upfront payments to secure supply, giving SK Hynix clear revenue visibility. AI investment is gradually expanding from initial compute power buildout to manufacturing optimization, energy management, and intelligent logistics. This broader shift is expected to continue driving demand for high-performance memory.
Goldman Sachs Report: AI Value Chain Closure and Sector Priorities
In its mid-2026 reports, Goldman Sachs systematically analyzed the structure and pricing dynamics of the full AI value chain. The core message is that AI competition has shifted from single links to complete closed-loop ecosystems. China accounts for roughly 16% of global AI revenue but receives significantly lower fund allocation, creating notable upside potential.
The report breaks the AI value chain into five prioritized tracks:
Power (Ranked #1 – Undervalued foundational bottleneck): AI training and inference consume massive electricity. Goldman forecasts China’s data center power demand to grow at a ~36% CAGR from 2025-2030. China’s advantages in supply scale, low-cost western green power, fast policy support, and construction speed stand out. Policies like East Data West Computing turn low electricity prices and land costs into real operational edges. Power equipment offers high rigidity and certainty.Semiconductors (Storage super-cycle with domestic substitution): Focus on DRAM, NAND, and HBM. AI server demand for these products is growing exponentially. Chinese storage players are advancing rapidly in production scale, cost-performance, and supply security. Recent revenue and export data already reflect price increases and share gains. This track offers strong profit elasticity and faster earnings realization.AI Infrastructure (Where capital expenditure actually lands): Covers servers, optical modules, liquid cooling, and data centers. China’s path is clear, with advantages in speed and cost-effectiveness. The report stresses that sustained spending comes from long-term inference and iteration rather than one-off training tasks.AI Models: China pursues low-cost, high-efficiency routes and shows strength in code, math, and multimodal areas. Goldman ranks this lower due to commercialization pace and competitive variables, requiring solid upstream infrastructure support.AI Applications (Lowest-risk monetization endpoint): China has the world’s largest single internet market. Massive real users and scenarios allow AI features to be directly embedded into existing products for monetization. Applications serve as the chain’s endpoint and driving force.
Goldman’s framework shifts capital allocation logic from betting on single links to betting on complete closed loops. Power and infrastructure provide certainty, semiconductors deliver elasticity, and models/applications offer longer-term excess returns. Overall, AI capex is expected to remain elevated, with greater weight placed on foundational bottleneck segments.
Post-Listing Institutional Participation
Following the debut, several institutions and product issuers joined or increased exposure:
Cornerstone Investors: Baillie Gifford Overseas, Coatue Management, and Situational Awareness Partners collectively showed interest in up to $7 billion of the ADRs.Underwriters: Bank of America, Citigroup, Goldman Sachs, and J.P. Morgan led the offering.ETFs and Derivatives: At least 10 fund managers, including Direxion and ProShares, filed to launch single-stock ETFs tracking SK Hynix. Leveraged and inverse products (SKHX and SKHZ) were also introduced shortly after listing.
The offering was oversubscribed more than 7 times, drawing broad interest from global long-only funds, tech-focused funds, sovereign wealth funds, and Asian-focused investors. This significantly expands U.S. institutional and retail access to SK Hynix.
SK Hynix’s Strategic Value and Risk Considerations
Within this AI value chain, SK Hynix sits at the core of semiconductor storage, especially HBM. Its technological leadership and long-term agreements align closely with the supply tightness and capex trends. The U.S. listing further opens international capital channels and strengthens its visibility in the global AI supply chain.
Geopolitical risks, macroeconomic fluctuations, and technological iteration remain factors to monitor.
Investment Outlook and Suggestions
In the short term, the U.S. listing improves SK Hynix’s global liquidity and visibility, with trading activity likely to stay elevated. Over the medium to long term, sustained AI capex and tight supply-demand dynamics provide solid support for its HBM business. Investors should track quarterly HBM shipments and capacity utilization.
For allocation, long-term investors may treat SK Hynix as a core AI theme holding. Those seeking diversification can supplement with related ETFs. Overall, decisions should align with personal risk tolerance and focus on the actual progress of AI infrastructure rollout.
SK Hynix’s U.S. listing marks a significant milestone for the company and reflects global capital markets’ emphasis on AI foundational technologies. Driven by both capital expenditure and technological progress, the memory sector demonstrates strong long-term resilience, with SK Hynix positioned at its center. Its future performance merits continued attention.
Disclaimer: This article is for informational purposes only and does not constitute investment advice. Markets are volatile; please conduct your own research.

I. Le cycle des taux d’intérêt qui a créé un nouveau marché
Si l’on remonte à 2021, le concept d’actifs du monde réel, ou RWA (Real-World Assets), relevait encore largement du domaine théorique. Il désignait le processus consistant à représenter des actifs issus de l’économie traditionnelle sur des blockchains et à permettre leur émission, leur détention et leur transfert sous forme tokenisée. À l’époque, toutefois, l’idée n’avait pas encore évolué vers un marché commercial convaincant. D’un côté, la Réserve fédérale avait maintenu les taux d’intérêt proches de zéro pendant des années, tandis que les rendements des bons du Trésor américain à court terme restaient inférieurs à 1 %, laissant très peu de revenus à faible risque dans la finance traditionnelle, suffisamment attractifs pour inciter à passer on-chain. De l’autre, l’industrie crypto se développait rapidement, et la DeFi, les NFT, le GameFi et d’autres secteurs spéculatifs généraient des retours spectaculaires. Les investisseurs étaient beaucoup plus intéressés par des actifs susceptibles de prendre dix fois, voire cent fois de la valeur, plutôt que par des produits offrant seulement quelques points de pourcentage de revenu annuel. Dans ces conditions, les RWA ressemblaient davantage à une orientation technologique qu’à une activité financière viable.

Le 8 juillet 2026, Binance Research a publié ce rapport approfondi intitulé Stablecoins : Transformer le paysage financier. Le rapport examine de manière systématique comment les stablecoins ont évolué, passant de simples passerelles au sein de l’écosystème crypto à une infrastructure centrale qui reconfigure le système financier mondial.
Cet article fournit une analyse détaillée des principaux enseignements du rapport, des données essentielles et des tendances émergentes, afin d’aider les lecteurs à comprendre les changements structurels en cours au sein des stablecoins et leurs implications profondes pour les particuliers, les institutions et l’ensemble du paysage financier.

À peine une semaine après son lancement en réseau principal public le 1er juillet, la valeur totale bloquée (TVL) de Robinhood Chain a dépassé la barre des 100 millions, culminant près de 106 millions avec une hausse sur 24 heures allant jusqu’à 159% à un moment. Cette croissance explosive, alimentée par des protocoles de prêt DeFi et amplifiée par la liquidité liée aux échanges de memecoins, a rapidement propulsé la nouvelle chaîne de couche 2 sous les projecteurs.
Le 8 juillet, une memecoin sur chaîne a brièvement vu sa capitalisation boursière dépasser 110 millions de dollars avant de revenir autour de 104 millions. Elle a offert plus de 13,9x de gains en 24 heures (encore plus à certains pics), avec un volume de transactions quotidien atteignant des centaines de millions de dollars. Le volume total sur 24 heures de la DEX de Robinhood Chain a bondi à plus de 500 millions de dollars, faisant de cette actualité l’un des sujets les plus brûlants du marché.

La publication des minutes de la réunion du FOMC de juin de la Réserve fédérale a détourné l’attention des investisseurs de la question familière « à partir de quand les baisses de taux commenceront-elles ? » vers un enjeu plus fondamental : qu’est-ce qui alimente l’inflation lors de la prochaine phase de l’économie américaine ?
Bien que les responsables politiques aient unanimement décidé de laisser les taux d’intérêt inchangés, les minutes font apparaître une inquiétude grandissante : les pressions inflationnistes évoluent plutôt qu’elles ne disparaissent. Au-delà des salaires et de la demande des consommateurs, les responsables de la Réserve fédérale ont de plus en plus mis en avant trois sources émergentes d’inflation persistante : les investissements liés à l’intelligence artificielle, la hausse des prix de l’énergie et l’augmentation des droits de douane.

En juillet 2026, Robinhood a dévoilé l’une des plus ambitieuses expansions produit de son histoire lors de son événement « The World is Flat » à Londres. La société a officiellement lancé Robinhood Chain, sa blockchain de couche 2 construite sur Arbitrum Orbit, en parallèle d’une suite de nouveaux produits, dont des actions tokenisées, un prêt décentralisé, des agents de trading alimentés par l’IA, des contrats à terme perpétuels, ainsi qu’une stratégie d’expansion mondiale accélérée.
À première vue, cela peut sembler être une autre plateforme financière qui lance sa propre blockchain. Mais en y regardant de plus près, Robinhood tente quelque chose de bien plus ambitieux : se transformer, non plus simplement en courtier en ligne, mais en couche d’infrastructure pour la prochaine génération de finance mondiale.

Introduction
Début juillet, des informations indiquant que Meta construirait une activité de cloud computing et préparerait la vente de capacité de calcul IA à des clients externes ont déclenché une réaction exceptionnellement vive dans l’ensemble du secteur de l’infrastructure IA. La réaction du marché a été étonnamment asymétrique : les actions de Meta ont fortement progressé, tandis que des sociétés de location de calcul IA comme CoreWeave et Nebius ont subi des pertes importantes. Dans le même temps, presque tout l’écosystème matériel IA — y compris AMD, Micron, SanDisk, ASML, TSMC, Samsung Electronics et SK hynix — a été soumis à une large pression vendeuse. À première vue, cela semblait n’être rien de plus qu’une entreprise technologique qui s’étend à une nouvelle ligne d’activité. En réalité, toutefois, ce que le marché évaluait n’était pas de savoir si Meta avait l’intention de commercialiser ses ressources GPU, mais si l’une des hypothèses fondamentales qui a soutenu l’industrie de l’IA au cours des deux dernières années commençait à changer.

Hier, une nouvelle annonce de stablecoin a rapidement dominé les discussions au sein de la communauté crypto ainsi que chez les investisseurs en actions américains. Plus de 140 entreprises et institutions ont conjointement lancé Open USD (OUSD), tandis que les actions de Circle (CRCL) ont immédiatement chuté d’environ 17,5 %. Dans le même temps, le dernier rééquilibrage de l’indice Russell a déclenché une pression vendeuse supplémentaire de la part des fonds passifs. Cet événement va au-delà du lancement d’un produit unique. Il marque l’accélération de l’intégration des stablecoins, issus d’outils natifs de la crypto, dans l’infrastructure grand public des paiements financiers, tout en poussant le marché à réévaluer l’impact concurrentiel réel des grands acteurs traditionnels de la finance qui entrent dans le secteur. Ci-dessous se trouve une analyse complète de l’événement, de l’OUSD et de ses implications sous plusieurs angles.

I. Pourquoi Grayscale a-t-elle tourné de nouveau son attention vers Solana ?
Au cours des dernières années, deux mots ont presque toujours défini Solana : la performance et les memecoins.
En tant que l’une des principales blockchains de couche 1 du cycle de marché précédent, Solana est montée en puissance grâce à son débit élevé, à ses faibles coûts de transaction et à une finalité quasi instantanée. Dans le même temps, des projets d’écosystème tels que BONK, dogwifhat (WIF) et Pump.fun ont fait de Solana le centre névralgique du boom des memecoins. Pourtant, cette perception a aussi éclipsé une transformation plus profonde en cours à l’échelle du réseau.

Fin juin 2026, Chainalysis a présenté un nouveau cadre appelé Blockchain Tracing Ontology, dans le but d’établir une manière plus standardisée et transparente de décrire l’intelligence liée à la blockchain. Plutôt que de lancer un autre produit d’analytique ou un nouvel outil d’enquête, l’entreprise tente quelque chose de bien plus fondamental : redéfinir la façon dont les données de traçage de la blockchain sont structurées, interprétées et communiquées.
Bien que le cadre soit encore au stade de proposition, il a déjà suscité un débat important dans l’ensemble de l’industrie des actifs numériques. À sa base, se trouve une question simple, mais aux répercussions considérables : l’intelligence de la blockchain a-t-elle besoin d’un langage commun ?

Le 26 juin 2026, OpenAI a officiellement présenté la famille GPT-5.6, dévoilant trois modèles distincts : Sol, Terra et Luna. Contrairement aux sorties précédentes centrées sur un seul modèle phare, GPT-5.6 marque un changement important dans la stratégie produit d’OpenAI. Au lieu de proposer un seul « meilleur » modèle, l’entreprise propose désormais un portefeuille complet de modèles conçu pour répondre à trois priorités différentes : une intelligence maximale, des performances équilibrées et une efficacité des coûts à haut débit.
Selon OpenAI, la série GPT-5.6 améliore considérablement les capacités en ingénierie logicielle, en informatique, dans le travail intellectuel professionnel, la recherche scientifique et la cybersécurité. Au lancement, les modèles ne sont disponibles que via une préversion limitée par l’API et Codex auprès d’un petit groupe de partenaires de confiance, avec une disponibilité plus large sur ChatGPT attendue ultérieurement.

Le 22 juin 2026, trade.xyz a officiellement lancé le contrat perpétuel ZHIPU-USDC sur le marché Hyperliquid HIP-3. Le contrat offre jusqu'à 10x de levier et permet le trading 24/7.
Ceci marque le deuxième actif coté à Hong Kong par trade.xyz. Le premier était MINIMAX (Xi Yu Technology / MiniMax Group, HK:0100), qui est devenu opérationnel le 18 juin 2026.
Aperçu rapide
Cet article commence par le contexte de l'entreprise ZHIPU et la percée technologique GLM-5.2, puis détaille la mécanique des contrats trade.xyz et leurs performances initiales, les compare avec MINIMAX, analyse plusieurs facteurs moteurs, explore l'écosystème Hyperliquid HIP-3, et enfin se penche sur le potentiel à long terme de la tarification on-chain pour des actifs divers.

Le 22 juin 2026, Marvell Technology est officiellement devenu un acteur de l'indice S&P 500. À première vue, cela peut sembler être un événement de rééquilibrage d'indice de routine. Cependant, lorsqu'on le considère à travers le prisme plus large du cycle d'investissement dans l'infrastructure de l'IA, la réévaluation de l'industrie des semi-conducteurs aux États-Unis et l'influence croissante des flux d'investissement passifs, l'inclusion de Marvell représente quelque chose de bien plus significatif. Cela constitue une reconnaissance formelle de la transformation réussie de l'entreprise, passant d'un fournisseur traditionnel de semi-conducteurs de communication à un acteur clé de l'écosystème d'infrastructure de l'IA. Pour Marvell, rejoindre le S&P 500 lui confère non seulement le statut de valeur refuge aux yeux des investisseurs mondiaux, mais augmente également les attentes concernant la croissance, la rentabilité et l'exécution à long terme. Ainsi, ce jalon est à la fois une réalisation et le début d'une phase plus exigeante de son parcours corporatif.