The current AI frenzy: featuring both "Tech and Internet" and "Real Estate" bubble characteristics, reminiscent of the 2008 global capital cycle

Twenty years ago, Loudoun County, Virginia, in the suburbs of Washington, D.C., was one of the hottest centers of the U.S. real estate bubble. It was subsequently hit hard by the global savings contraction and the 2008 subprime mortgage defaults. Two decades later, the area has become blanketed with data centers, attracting capital from around the world—including Canadian pension funds and Japan’s SoftBank—to build large-scale AI infrastructure, making it one of the globe’s most densely concentrated nodes for internet and AI computing power.

By the 2026 fiscal year, over 40% of the county's tax revenue ($1.2 billion) will come directly from data centers.

This extreme reliance on a single asset has caused local officials to once again feel the fear present on the eve of the 2008 crisis—Loudoun County has not only recently halted the approval of new data centers, but also established a $120 million “fiscal stabilization fund” and required operators to pay liquidation guarantees, in case the computing power bubble bursts and the county faces another financial collapse like before.

Loudoun County's anxiety is a microcosm of the global macroeconomic capital imbalance.

"Data Center Alley" in Ashburn, Loudoun County, video screenshot

The International Monetary Fund (IMF) forecasts that from 2023 to the end of 2026, the global cumulative surplus savings will reach $1.95 trillion, about one-third higher than the increase in the three years before the 2008 crisis.

The asset pool absorbing these funds has shifted entirely from real estate to the AI industry chain.

Global Savings Shuffle: From Buying U.S. Treasuries to Betting on U.S. Equities and AI

This time, the flow of global capital has undergone a key restructuring:

• The channel has shifted from official to private: In the early 2000s, overseas savings in Asia and other regions were mainly managed by foreign central banks, buying large amounts of U.S. Treasuries, which objectively suppressed global borrowing costs. Today, however, massive amounts of capital are controlled by overseas private funds seeking high returns, flowing directly into U.S. equities and tech company bonds.

• Risk exposure is highly concentrated: Bloomberg Economics analysis shows that by the end of June 2026, overseas investors' exposure to core AI-related industries in the U.S. stock market will reach about $4.9 trillion (roughly equivalent to Germany’s annual GDP), with their proportion of all U.S. stock holdings soaring from about 10% in 2019 to nearly 20%.

Bloomberg report screenshot

Hung Tran, a senior research fellow at the Atlantic Council, commented: “It’s like the petrodollars of the Middle East in the 1970s. After Asian tech companies earn dollar profits, they cycle them back into the U.S. market. Today’s ‘tech dollars’ have become a new version of petrodollars.”

Frederic Neumann, HSBC’s Chief Economist for Asia-Pacific, stated: “Asian capital is the fuel feeding the U.S. AI boom. This is entirely built upon fragile global demand and capital flow imbalances. Should AI investments falter, the U.S., the dollar, and global markets will all face painful adjustments.”

Both “Tech” and “Real Estate”: The Structural Mutation of AI Assets

Why can AI absorb such massive cross-border capital? CICC’s research report “The American AI Era of ‘Marshmallows’” puts forward a core logic: The U.S. AI industry simultaneously possesses the dual attributes of hard-core technology and ‘quasi-real estate’.

From a technological perspective, AI’s per capita efficiency and its capacity to replace labor surpass the internet companies of the past.

Data shows that since ChatGPT’s release in 2022, nonfarm employment in U.S. telecommunications and IT industries has dropped by 6%. In contrast, during the internet bubble from 1995 to 2000, employment in these sectors surged by over 60%. If you measure labor intensity by the employee-to-fixed-assets ratio, the labor density index of Google, Microsoft, and Meta tumbled from 100 in 2021 to just 37 by 2025.

Yet from a balance sheet perspective, AI looks more like a heavy-asset investment game.

CICC statistics show that by the second quarter of 2026, the fixed assets of top five players—Microsoft, Alphabet, Amazon, Meta, and Oracle—account for more than 40% of total assets on average (traditional heavy-asset industrials average 37%). At the peak of the 2000 internet bubble, tech companies like Microsoft and Cisco had an average fixed assets ratio of only 15%.

Columbia Business School economist Stijn Van Nieuwerburgh estimates that by 2032, cumulative global investment in AI infrastructure will reach nearly $9 trillion. As a share of GDP, this will surpass the railroad boom of the 19th century, making it the largest capital expenditure expansion in U.S. history.

Massive investment in fixed assets gives AI enterprises a “collateral property” akin to real estate, unlocking a credit financing logic very different from the dot-com era.

Debt and Securitization Surge: Goodbye VC, Hello Subprime-era Financial Engineering

Asset characteristics determine financing modes.

CICC’s report points out that the 1990s internet bubble mainly relied on VC and stock IPO (equity financing), while current AI investments, with their massive accumulation of data centers, land, and equipment, are rapidly moving toward the debt expansion and structured financial models seen during the 2000s real estate bubble.

As of the second quarter of 2026, representative AI companies have a net cash flow from financing that is 16.2% of revenue; net debt financing accounts for as much as 21.2%, while equity financing is now in net buyback mode.

This takes shape in three major financing pillars:

1. Massive corporate bond issuance: According to Dealogic, in just the first six months of 2026, five AI giants issued $159 billion in bonds globally, compared with just $17 billion for the whole of 2024.

2. Off-balance sheet debt and private credit expansion: According to Bloomberg, the top five giants have cumulatively signed nearly $3 trillion in long-term leases, procurement commitments, and guarantees that are not reflected on their balance sheets. The Bank for International Settlements (BIS) estimates that large private credit institutions have issued over $200 billion in direct loans to AI-related firms, accounting for nearly 8% of their total, effectively replacing traditional bank lending.

3. Computation asset securitization (ABS): Mirroring the mortgage-backed securities (MBS) of old, computing power leases are now being packaged as ABS assets. Following CyrusOne, Sabey, Stack, and QTS, in early 2026 Compass Datacenters issued $830 million in data center ABS, with underlying assets directly tied to long-term cloud computing leases, distributed to pension and insurance funds seeking fixed income.

Comparing Historical Bubble Bursts: Where Are the Cracks Appearing?

Looking back, CICC research notes that the eventual bursting of both the dot-com and real estate bubbles coincided with three phenomena: macroeconomic and sector high leverage, deteriorating fundamentals, and systemic tightening of financing conditions.

Currently, interest coverage ratios for AI companies are still high (Google about 82 times, Oracle about 7 times), the sector’s nominal profits remain positive, and the systemic conditions for a bubble burst are not yet fully mature.

David Chao, Global Market Strategist at Invesco, said: “The technology is real, the applications are real, and many leading companies are highly profitable. What I worry about is that investor expectations will soon become impossible to satisfy.”

However, short-term marginal cracks in capital flows can no longer be ignored:

• Rapidly rising debt costs: In July 2026, Meta’s $12 billion bond had a coupon rate of 7.5%, 0.4 percentage points higher than at the end of 2025; CoreWeave’s high-yield bonds saw yields rise by more than 0.5 percentage points after the Fed hiked rates in September; S&P recently downgraded Oracle to BBB-.

• Government fiscal buffers exhausted: Former IMF Chief Economist Pierre-Olivier Gourinchas stated bluntly: “Even a technological revolution is not a blank check with unlimited credit.” He warned that during the dot-com bubble, public debt as a share of GDP was only one-third to half of today’s level. Now, U.S. debt exceeding $40 trillion and still-elevating Treasury yields have severely eroded fiscal flexibility to face crises.

• Private institution exit risk: Economist Gian Maria Milesi-Ferretti emphasized that many hedge funds are levering up on U.S. Treasuries as collateral to play the AI game: “These institutions rebalance very rapidly. Today’s risk isn’t foreign central banks selling Treasuries, but these yield-seeking private institutions being forced to dump holdings to compensate for losses elsewhere.”

From physical data centers facing moratoriums in Loudoun County to the layers of computation ABS on Wall Street, this financial game, blending tech valuations with real estate debt models, is drawing global capital into an unprecedented network of resonance.