Bank Data Center Loans Hit Exposure Limits, Agricultural Credit Resells HKD Loans
Miles Bennett
Crédit Agricole CIB is reselling a HK$150 million loan stake after hitting its internal data-center exposure cap — a sign that banks are running into structural lending limits as AI-infrastructure debt surges past US$334.5 billion this year alone.
What is this loan about?
In 2023, Crédit Agricole CIB joined other banks to lend HK$1.6 billion to ESR Group for converting a cold-storage facility in Kwai Chung, Hong Kong, into a data center.
The bank is now reselling roughly HK$150 million of that loan because it has hit its own internal lending cap for the data-center sector.
This means → the loan itself is not in trouble; the bank's portfolio-wide exposure to data centers has simply maxed out.
Is this happening at other banks too?
Morgan Stanley previously explored a significant risk transfer (SRT) — packaging data-center loan risk and selling it to investors — to free up lending capacity.
Asset manager Voya Financial has already started capping its exposure to long-term contracts tied to Big Tech's AI infrastructure.
In plain terms = from lenders to asset managers, financial institutions are finding their data-center debt exposure too concentrated and are actively de-risking.
How big is the data-center funding wave?
Moody's expects at least US$3 trillion to flow into data-center investment over the next five years, mostly debt-funded.
Year-to-date, at least US$334.5 billion in bonds and loans have been issued for AI infrastructure — nearly double the US$185.5 billion raised in all of 2025.
In Asia-Pacific, DayOne Data Centers is negotiating a US$7 billion loan that would be the largest single data-center financing in the region.
What does this mean for the market?
Meta, Alphabet, and other tech giants continue to borrow aggressively for AI infrastructure, yet doubts are growing about whether these investments will yield stable returns.
Banks tightening their exposure caps = a structural supply-side constraint on data-center debt; borrowing is no longer as easy as it was.
This reflects a critical test for the AI-infrastructure boom: if banks keep pulling back, financing costs for future projects will most likely rise.
Content is for reference only, not financial advice.