# Taiwan Central Bank Names AI Returns and Financing as a Global Outlook Risk.

By John Kim

Canonical URL: https://www.tokenpost.com/news/business/41569
Published: 2026-10-27T06:18:55.000Z
Updated: 2026-10-27T06:18:55.000Z
Section: Business

> Its Oct. 27 materials identified data-center bottlenecks and complex financing as channels for possible spillovers to private credit and banks if AI returns fall short.

Taiwan's central bank identified uncertainty over AI investment returns and rising financing risks as one of the major risks to the global economic outlook, with data-center financing a potential channel for financial stress.

The bank submitted written materials Tuesday, Oct. 27, ahead of its governor's scheduled appearance before the legislature's Finance Committee Wednesday. U.S. data-center projects face constraints from power availability, equipment supply, community opposition and local government reviews, which could delay openings. Newer cloud providers and data-center developers are relying more on private credit, asset-backed financing and special-purpose vehicles (SPVs), raising leverage and adding complexity to their funding.

Capital spending by Alphabet, Amazon, Microsoft and Meta was projected at $715.5 billion in 2026, up 90.3% from 2025, with spending projected to reach $991.6 billion in 2027. Operating cash flow has not kept pace with that spending, reducing free cash flow and increasing the companies' need for outside financing.

Global data-center financing needs were estimated at about $3.2 trillion from 2026 through 2028, with $700 billion projected from private credit and $650 billion from investment-grade bonds. The four cloud providers' off-balance-sheet commitments and liabilities totaled about $2.424 trillion, roughly four times the $604 billion in long-term borrowings and existing lease liabilities on their balance sheets. These included more than $1.52 trillion in commitments to purchase chips and about $904 billion in future leases for data centers under construction, payable after the centers are completed.

There is also a mismatch between chip life spans and financing terms. GPU accounting depreciation runs four to six years, while chip life cycles have shortened to two to three years; SPV leases and debt typically run five to 15 years. If cloud companies' revenue weakens, spending cuts or defaults could transmit risk from the AI supply chain to nonbank lenders and, if those lenders borrow from banks, to the banking system.

Changes to U.S. cloud providers' capital spending would affect Taiwan's AI hardware export growth. The central bank's September forecast put Taiwan's 2026 economic growth at 11.48%, with net external demand and capital formation together accounting for 75.6% of projected growth. The central bank will continue monitoring AI prospects and global financial conditions as it adjusts policy to maintain price and financial stability.

The governor is scheduled to appear before the Finance Committee Wednesday, Oct. 28.