When 90% of global liquidity flows to just five exchanges, markets like Hong Kong and India are being left behind.

We are now past the halfway point of 2026. If one had to summarize the defining feature of the year in global capital markets, it would be this: this is not a broad-based global bull market, but a targeted, intense, and unprecedented capital migration effect.

What is this migration effect? It means capital is not flowing evenly across the world but is being systematically drawn into a few specific "low-lying" areas. In other markets left behind, the water level is receding.

A single data point illustrates the intensity of this shift. Since the start of 2026, among exchanges with a market capitalization of over one trillion US dollars, the increase in value has been concentrated in just five places: the US Nasdaq, the Korea Exchange, the Taiwan Stock Exchange, the Tokyo Stock Exchange, and the Shanghai and Shenzhen exchanges. Their net increases are $6.2 trillion, $2.19 trillion, $1.56 trillion, $1.06 trillion, and $1.06 trillion, respectively.

Combined, these five exchanges have absorbed over $12 trillion in market value growth this year. Other global markets have seen either negligible gains or outright shrinkage.

Extending the timeline back to the end of 2024 only reinforces this conclusion. Since late 2024, the Nasdaq has added $13.13 trillion, the Shanghai and Shenzhen exchanges $5.2 trillion, South Korea $3.39 trillion, Taiwan $2.3 trillion, and Japan $2.36 trillion. Even over this longer period, the Hong Kong exchange's increase was only $1.46 trillion, and India's performance was even weaker.

Looking closely, one detail is particularly striking. The Korea Exchange's market cap increase this year is $2.19 trillion, ranking second globally, just behind the Nasdaq. Taiwan's $1.56 trillion ranks third. These two economies, which together account for less than 3% of global GDP, are outperforming Europe, Japan, and Hong Kong in stock market growth.

Why?

The answer points in one direction: the geographical concentration of the AI supply chain.

If you break down the upstream segments of the global AI supply chain, you find something interesting. While AI chip design is in the US and AI computing power deployment is in the US and China, the manufacturing of AI hardware is highly concentrated in East Asia: South Korea's HBM memory chips, Taiwan's advanced semiconductor wafer fabrication, Japan's NAND flash and semiconductor materials. The growth in global stock markets is precisely flowing along the geographical distribution of this supply chain, from downstream to upstream.

Core Insight

The 2026 global capital market is not experiencing a "broad rally" but a "re-pricing"—a valuation of the strategic importance of every link in the AI supply chain. Markets not on this chain, regardless of their GDP size or cheap valuations, struggle to attract incremental capital. The clear underperformance of Hong Kong's Hang Seng Index compared to South Korea, Japan, and Taiwan this year, alongside the weakness of India's Nifty, is the most direct evidence.

However, the details are more fascinating than one might imagine. Even within the AI supply chain, the "migration patterns" differ vastly between markets. South Korea's surge is propped up by two memory chip giants, Japan's turnaround is led by a once-discarded business unit, and Taiwan's story involves a leader followed by a full supporting cast.

Let's examine each one.

The South Korean Duality

On June 22, 2026, a momentous event occurred in South Korean financial history: SK Hynix's market capitalization briefly surpassed that of Samsung Electronics, marking the first change at the top of South Korea's corporate value ranking in 25 years and 7 months.

This is not a simple ranking shift. Samsung Electronics has been the undisputed leader since 1999, representing not just a company but half a century of South Korea's industrial achievement. Meanwhile, SK Hynix, a company that was nearly bankrupt two decades ago, has staged a comeback powered by one thing: HBM (High Bandwidth Memory), the core memory chip for AI data centers.

As of June 18, 2026, Samsung's market cap was 2,069.58 trillion won (approx. $1.37 trillion), and SK Hynix's was 1,969.91 trillion won (approx. $1.31 trillion), a difference of less than 5%. The brief overtaking on June 22 was a powerful signal.

The more critical figure is this: these two companies together account for 52.1% of the total market capitalization of the Korea Exchange. In 2019, this figure was 20.56%. By the end of 2024, it was 30.24%. It has surged from 20% to 52% in less than three years.

Year-to-date in 2026, SK Hynix's market cap has grown 315.65%, and Samsung's has grown 191.59%. Excluding these two, the rest of the South Korean stock market has risen only 35.99% collectively.

This data illustrates one point: investing in the South Korean market now is essentially a bet on two memory chip stocks. Missing Samsung and SK Hynix means missing this rally.

SK Hynix is currently the absolute leader in the global HBM market, serving as the almost exclusive supplier of HBM3E for NVIDIA's H200 and B200 accelerator cards. Samsung, while catching up, has lagged in HBM3E yield, missing out on the peak红利 of the recent AI server deployment cycle.

The core logic here is not about "how much the Korean market has risen" but about the structure of the rise. From 2019 to 2026, the market's total value has expanded, but this expansion has been unevenly "pulled up" by the explosive growth of these two chipmakers.

This raises a fundamental question: Is it good or bad for a national stock market to be dominated by two companies? In the short term, it's beneficial, creating wealth as seen in South Korea's $2.19 trillion growth. In the long term, it means market volatility is entirely tied to the HBM and DRAM cycle. A slowdown in AI capital expenditure or a breakthrough by competitors could trigger a sharper correction than in any other market.

The Japanese Phoenix

Japan's story is equally compelling but follows a different script: a business unit sold off as a "discard" has, four years later, become the king of the Japanese stock market.

This company is KIOXIA. It listed on the Tokyo Stock Exchange in December 2024 with an IPO market cap of 0.88 trillion yen. By the end of 2025, its市值 had grown to 5.65 trillion yen—a sixfold increase in a year, a feat in Japan's typically stagnant market. The drama accelerated in 2026. By June 10, its市值 had surpassed that of Toyota Motor, reaching 59.43 trillion yen by June 18—a year-to-date surge of over 950%, rocketing from 169th place to number one.

KIOXIA's story is dramatic. Its predecessor was Toshiba Memory. In 2018, Toshiba Group, reeling from massive losses in its US nuclear business, was forced to sell its most valuable asset—the world's second-largest NAND flash maker. A consortium led by Bain Capital acquired it for about $18 billion, renaming it KIOXIA in October 2019.

Then AI arrived. Training and running large language models require massive data storage, and NAND flash is a core component of AI data center storage architecture. By the second half of 2025, NAND prices had risen over 300% from their lows. KIOXIA's latest quarterly report showed operating profit surpassing that of Toyota Motor during the same period.

KIOXIA, focused solely on NAND flash, holds about a 14% global market share. Its key technology is BiCS Flash, a 3D stacked NAND developed with Western Digital.

It's important to note that Japan's AI rally isn't a one-company show. Companies like Tokyo Electron, Murata Manufacturing, Panasonic, and Sumitomo Electric Industries have also seen市值 gains exceeding 100% this year. Japan's AI boom is characterized by a broad-based rally across upstream materials and components, contrasting with South Korea's concentrated "duet." This reflects their different roles: South Korea is a powerhouse in memory and logic chips, while Japan is the "upstream king" of materials and equipment.

The Taiwanese Ecosystem

Taiwan's story is different from both South Korea and Japan. While South Korea is highly concentrated and Japan has a star leader with broad support, Taiwan features a flagship company leading an entire army.

First, Taiwan Semiconductor Manufacturing (TSM). As of June 18, 2026, TSMC's market cap reached NT$76.34 trillion (approx. $2.38 trillion), accounting for 42.87% of the total market value of the Taiwan Stock Exchange. This level of concentration is rare globally.

However, a surprising data point is that TSMC's year-to-date gain is 53.74%, while the rest of the Taiwan stock market (excluding TSMC) has risen 62.94% collectively. The "leader" is actually underperforming its followers.

Companies like MediaTek, Delta Electronics, ASE Technology Holding, Unimicron Technology, and Accton Technology have all posted gains exceeding TSMC's, ranging from 62% to 82%.

This decentralized performance structure reflects the comprehensiveness of Taiwan's AI supply chain. TSMC handles advanced wafer foundry work. But an AI server is a complex system requiring massive配套需求: MediaTek's edge AI chips, Delta's power and thermal management, ASE's advanced packaging, Unimicron's IC substrates, Accton's data center switches.

The underlying logic of Taiwan's AI bull market is not "one exceptionally strong company" but the fact that an entire AI hardware supply chain is clustered on the island. This industrial agglomeration effect is unmatched globally. While South Korea ranks second in市值增量 this year, Taiwan exhibits the healthiest market in terms of rally "breadth."

A pattern is clear: South Korea, Japan, and Taiwan, all deeply embedded in the AI supply chain, are experiencing a "re-pricing by AI." The core driving force for this process comes from the world's largest AI buyer—the United States.

The US Reallocation

The US stock market is the starting point of the global AI capital migration—all funds ultimately converge here. But it is a mistake to think the US market is experiencing a "broad rally." Examining the performance of the top 20 US companies reveals a harsh truth: intense differentiation is occurring even within the AI sector, while non-AI sectors are being systematically abandoned.

Year-to-date, Micron Technology (MU) is up 298.11%, Intel (INTC) is up 265.24%, Advanced Micro Devices (AMD) is up over 100%, and ASML Holding NV (ASML) is up 80.37%. Their commonality? They are all core suppliers in the AI hardware supply chain.

TSMC's US-listed shares are up 52.07%, Broadcom is up 19.26%, Alphabet (Google) is up 17.64%, and NVIDIA (NVDA) is up 12.51%. These are still AI plays, but their gains have moderated due to their massive base. NVIDIA's 12.51% gain corresponds to a市值增量 equivalent to adding an entire Intel.

Most striking are the declines on the other side: Microsoft (MSFT) is down 10.2%, Tesla is down 8.3%, Meta is down 5.8%, Berkshire Hathaway is down 4.2%, JPMorgan Chase is down 3.5%.

This illustrates a crucial trend: the "headline effect" of the AI rally is migrating to an "upstream effect." In 2023-2024, leaders were downstream players like NVIDIA, Microsoft, and Google. By 2025-2026, the focus shifted from "who will profit from AI" to "who is building AI infrastructure." "Shovel suppliers" like Micron, Intel, and AMD have become the biggest winners.

This is not short-term speculation but an industrial infrastructure重构 on a decadal scale, fueled by annual capital expenditures of hundreds of billions of dollars on AI computing, which largely translate into revenue and profit for AI hardware suppliers.

But where is all this money coming from? The answer extends far beyond the stock market.

The Debt Market Channel

While everyone watches the stock market, US tech giants have been quietly siphoning massive liquidity from the global bond market.

In 2025, Amazon, Alphabet, Meta, Oracle, and Microsoft collectively issued $121 billion in bonds—a historical record, quadruple their average annual issuance from 2020-2024. In just the first five months of 2026, these five giants, plus newcomers NVIDIA and SpaceX, have issued $159 billion, already exceeding the full-year 2025 total.

Why are tech giants issuing debt instead of equity? Two key reasons: First, financing costs. Their bond issuance rates are far lower than their cost of equity. Second, and more crucially, debt does not dilute existing shareholders' equity. Issuing long-term, low-interest debt is the optimal solution: borrow global investors' money to build AI infrastructure and keep all the future returns.

This is the "second leg" of the AI capital migration: equity markets吸走 stock investors' money, and bond markets吸走 fixed-income investors' money. These two channels work in tandem to funnel global capital into every link of the AI supply chain.

Underlying Driver

The combined AI capital expenditure of four tech giants (Amazon, Alphabet, Meta, Microsoft) is projected to reach $610-660 billion in 2026, a year-on-year increase exceeding 60%. A significant portion of this spending becomes orders for NVIDIA GPUs, wafer purchases from TSMC, prepayments for SK Hynix's HBM, and long-term contracts for KIOXIA's NAND flash. The市值增量 in global stock markets—whether Nasdaq, South Korea, Taiwan, or Japan—is fundamentally driven by this astronomical capital expenditure list.

This "migration effect" is not merely a securities market phenomenon; it is the financial expression of a global AI infrastructure race. As long as tech giants continue investing in AI at a rate exceeding $600 billion annually, capital will continue to flow along the supply chain map to South Korea, Taiwan, and Japan, inflating their market values further.

When Will It End?

When might this AI feast conclude? From a fundamental perspective, it has support. This is not pure liquidity-driven speculation. It is backed by real demand for AI models, expanding tech giant capex, and genuine semiconductor supply tightness. SK Hynix's HBM capacity is reportedly booked by NVIDIA through 2027; KIOXIA has stated its 2026 NAND capacity is sold out with tight supply lasting at least until 2027; TSMC's 3nm/5nm lines are running at full capacity.

However, having support does not mean there are no risks. Three major risk points warrant vigilance:

First, the pace of AI application adoption and monetization. If user growth and commercial returns consistently fall short of expectations in the next year or two, the tech giants' massive annual capex could face pressure to shrink, potentially triggering a severe "inventory correction" across the AI hardware supply chain, with South Korean, Taiwanese, and Japanese markets bearing the initial brunt.

Second, geopolitics. Can SK Hynix continue selling HBM to China? Will TSMC's advanced processes face further US restrictions? Will KIOXIA's Japanese factories benefit from a "diversification effect" due to US-China tech decoupling? Major shifts in these variables could profoundly impact the AI supply chain's geographical distribution.

Third, financing conditions. The前提 for tech giants' massive low-cost debt issuance is a relatively宽松 global interest rate environment. A resurgence of inflation forcing the Fed to resume rate hikes could significantly increase borrowing costs, stalling the "financing engine" for AI capex.

For now, these risks remain possibilities. Based on current data, there is no immediate reason for this AI feast to stop on its own. A bubble supported by strong fundamentals is the hardest kind to burst—because it doesn't feel like a bubble at all.

A final question emerges: Could the current格局, where over 90% of global stock market gains are absorbed by four or five "AI-focused markets," become the new normal for the next five or even ten years? If so, investors in markets outside this chain—Hong Kong, India, Europe, South America—are not just missing a bull market but confronting a structural shift of an era.