In recent years, generative AI has triggered an unprecedented wave of investment in computing power and infrastructure. From GPU chips to data centers and power infrastructure, the capital intensity of the AI industry has risen sharply, forming what may be described as a rare technology infrastructure capital cycle. International Data Corporation (IDC) estimates that the global AI infrastructure market will reach USD 758 billion by 2029. Some reports even suggest that without the combined USD 344 billion in capital expenditures from the “gang of four” technology giants Microsoft, Amazon, Meta, and Alphabet, overall US economic growth could slip into recession, highlighting the critical role that AI investment currently plays in the macroeconomy. With unprecedented capital volumes, the defining feature of this technology cycle is not merely innovation, but a fundamental shift in the sources and flow structures of capital.

The speed and scale of this investment are already challenging traditional financial regulatory frameworks, but under the regulatory and capital adequacy requirements of Basel III, participation in the highly volatile and capital-intensive AI industry by the traditional banking system is strictly constrained. Bank capital has therefore begun to flow outward on a large scale, shifting toward private equity funds, sovereign wealth funds, and internal corporate financing chains. Within this structure, private credit and structured funds are rapidly replacing bank lending as the new engines of financing. At the same time, a phenomenon of internal financing has emerged within the AI industry chain, forming a closed loop of ecosystem finance. As a Financial Times editorial has warned, when private credit is packaged by private funds and resold to insurers, risk may temporarily leave the banking system, but can be redistributed and accumulate implicitly off balance sheets through complex leverage structures. This capital spillover phenomenon is reshaping the multi trillion-dollar global AI investment landscape at remarkable speed.

The strict capital adequacy and liquidity requirements of Basel III have forced banks into sovereign bonds and high-grade short-term securities. Instead, tech firms have turned to non-bank financial intermediaries including private equity, family offices, hedge funds, and internal corporate financing.

This development marks a structural shift in capital flows and financial intermediation. Capital flows are moving from a credit driven model to a capital driven model: the AI industry no longer relies primarily on bank lending for indirect financing, but rather equity, notes, supply chain financing, and tokenized assets. Meanwhile, GPU leasing contracts, and AI model revenue streams are increasingly becoming collateral or tradable investment assets for direct financing.

The rapid expansion of the AI industry is rewriting risk transmission in new ways. One of the most widely discussed topics in capital markets recently is the so-called AI perpetual engine – which refers to a circular investment and long-term procurement model between technology giants such as OpenAI, NVIDIA, and Oracle, supported by financing from SoftBank’s international private funds and sovereign wealth funds in the Middle East. This model forms a powerful mechanism for capital accumulation. Many AI infrastructure companies finance investment through advance payments, supply chain loans, equipment sale leaseback arrangements, and structured notes, often without going through the on-balance sheet supervision of banks.

Because the non-bank financial intermediary system lacks unified disclosure standards and centralized clearing, leverage levels and asset quality remain opaque. If the business cycle reverses, risk may quickly spread back into the market through inter-corporate debt relationships and leveraged private funds. For example, large US data center operators have supported multi-billion dollar GPU purchases with short-term private bonds, then packaged AI leasing revenues into high-yield products to be resold to insurers or pension funds. This multilayered AI credit chain can improve capital efficiency during periods of expansion, but may trigger cascading disruptions if market valuations change or liquidity tightens. In particular, private credit products often involve complex cross-collateralization and hidden covenants, making asset valuation and liquidation far more complex than in public markets.

This all comes at the same time that stablecoins and real world asset (RWA) tokenization are reshaping the global financial settlement layer. Initially serving merely as trading media in cryptocurrency markets, stablecoins have now evolved into a new form of on-chain dollar reserve currency. The daily transaction volume of stablecoins such as USDC and USDT has surpassed that of payment networks such as Visa and PayPal, as well as certain sovereign currency payment systems. Widely used in cross-border trade, supply chain settlement, and capital flows in emerging markets, their digital settlement systems bypass the traditional SWIFT payment network and provide lower-cost and more efficient T+0 settlement. Particularly in regions with strict capital controls, stablecoins have become effective tools for B2B payment and liquidity management. According to the latest data from a16z, more than half of the USD 30 billion currently issued for RWA originates from private credit – confirming that private credit is increasingly bypassing the banking system and raising on-chain capital through stablecoins.

RWA transforms traditional assets such as bonds, funds, and receivables into tokenized forms that can be settled and fractionally traded on global blockchain networks. This represents not only technological innovation, but also an evolution in financial system – reflected in asset standardization, real time settlement, and regulatory transparency. The direct circulation of tokenized bonds and fund shares in global markets enables synchronized settlement of capital and assets on the blockchain, shortening counterparty risk cycles, and allowing reserves, underlying assets, and redemption activities to be tracked, reducing information asymmetry.

The BUIDL fund token launched in March 2024 by BlackRock, in cooperation with the tokenization platform Securitize, a US RWA platform regulated by the US Securities and Exchange Commission and Financial Industry Regulatory Authority, represents one of the best examples of RWA implementation. The fund allows investors to subscribe to a short-term US Treasury market fund using the USDC stablecoin. BUIDL is recorded on Ethereum mainnet, and issued and transferred by Securitize, and the underlying assets are held and administered by Bank of New York Mellon. By integrating a tokenized fund into the existing legal framework, this structure not only achieves regulatory alignment, but also introduces settlement innovation by allowing investors to enter and exit the fund around the clock using stablecoins. Most importantly, the architecture shifts the core functions of fund registration, settlement, and reserve verification entirely onto the blockchain.

The success of BUIDL symbolizes the formal integration of traditional finance and blockchain technology. Banks no longer dominate capital supply, but instead function as trust anchors behind RWA structures. In the future, when AI companies, supply chain firms, and even governments can issue debt through tokenized structures, the center of gravity of financial markets may shift from on balance sheet assets and liabilities to on-chain settlement.

This transformation provides financial regulators with a new reference model. The convergence of AI with RWA is gradually shaping the concept that digital assets themselves represent capital markets. Power leasing rights for AI computing, model revenue rights, and data center debt claims can all serve as underlying assets. Traditional securitization struggles to accommodate the highly heterogeneous and rapidly fluctuating values of AI assets, but RWA structures can use smart contracts to enable fractional ownership and programmable revenue distribution for more efficient capital formation and asset liquidity.

Driven by stablecoin and blockchain development, the traditional bank function of financial intermediation is being disassembled and rebuilt into three roles: settlement infrastructure, custodianship, and regulatory trust. Banks ensure the bridge between fiat currency and on-chain assets, manage custody of assets, RWA tokens, and stablecoin reserves, and provide regulatory trust services such as compliance verification, KYC, and auditing.

In this model, banks increasingly provide the foundational infrastructure that supports financial order, but this transformation brings three profound changes. First, funds cross regulatory boundaries and flow directly into AI industry chains, de-territorializing capital formation. Second, traditional securitization markets are being partially replaced by blockchain-based programmable liquidity pools, redistributing market liquidity. Third, with the modular reconfiguration of banking functions, with lending, custody, settlement, and risk management distributed across different on-chain protocols, the financial system is no longer organized primarily around institutions but around protocols.

Stablecoins represent a financial revolution without borders, licenses, or dependence on national sovereignty. Taiwan’s banking system is well-capitalized and highly liquid, yet it remains largely an observer in the global competition surrounding RWA and stablecoins. The Monetary Authority of Singapore has advanced Project Guardian into its second phase, testing the use of multi-currency stablecoins and tokenized assets in cross-border payments, in contrast, creating substantial regional competitive pressure for Taiwan.

Facing the dual wave of AI and Web3, Taiwan’s financial industry must act if it intends to maintain regional competitiveness. First, regulators could promote pilot programs for on-chain settlement and regulatory experimentation under the sandbox framework of the Financial Supervisory Commission. Domestic banks could serve as custodians, while working with blockchain companies to validate RWA tokenization structures. This could further support the development of RWA trust and custody services, positioning banks as the fiat trust layer within on-chain capital markets. Such initiatives would provide Taiwan’s high-tech supply chains with new international financing channels, while reinforcing Taiwan’s strategic role as a regional technology capital hub. If banks can position themselves as trust intermediaries and regulatory agents within stablecoin and RWA structures, they may regain strategic influence in the emerging digital capital markets.

The financial order of the future will no longer revolve around bank balance sheets, but around on-chain liquidity pools and regulatory trust layers. Those who master the language of tokenized finance and control settlement mechanisms will shape the global AI capital competition. Banks must transform from passive intermediaries into guardians of on-chain financial order and builders of trust if they are to retain their central role in the new era.

The author is a co-founder of TokenBacon Taiwan Co., Ltd.