← Back to articles
← Previous (older)
Beijing Just Killed a $2bn Meta Deal — and Sent Silicon Valley a Message
Next (newer) →
Why the Pentagon Just Froze 165 American Wind Farms
finance · technology · May 04, 2026

Wall Street's AI Bottleneck: Why Banks Are Choking on Data Centre Debt

No reader ratings yet.
Log in to rate this article
📰 Reading Passage

When you ask an AI chatbot a question, the answer is computed inside an enormous, power-hungry warehouse stuffed with specialised chips. Someone had to pay to build that warehouse — usually by borrowing from a bank. And right now, according to a Financial Times report, the banks doing that lending are running out of room.

Groups including JPMorgan Chase, Morgan Stanley, SMBC and MUFG are scrambling to find new investors willing to take slices of the loans they've made to data centre projects. The reason is the sheer scale of the AI build-out. Oracle, the database company that has reinvented itself as an AI infrastructure giant, and CoreWeave, a specialist cloud provider, have together borrowed hundreds of billions of dollars to construct facilities across the United States. One package alone — roughly $38 billion of construction debt tied to Oracle-leased sites in Texas and Wisconsin — would rank among the largest debt deals in history. Matthew Moniot, a credit executive at Man Group, told the FT that 'the sizes we're talking about' are unlike anything lenders have considered before, and that 'banks very quickly start choking.'

The choking metaphor is precise. Every bank operates under internal rules and government regulations that cap how much exposure it can have to a single borrower. Once Oracle shows up as the tenant or guarantor on dozens of separate financings, lenders bump into those caps and cannot take on more — even if every individual loan looks safe. So banks are turning to a workaround that has spread from Europe to North America in recent years: the Significant Risk Transfer, or SRT. In an SRT, the bank keeps the loan on its books but sells the *risk* of default to outside investors — typically private credit funds and insurers — in exchange for a fee. That frees up the bank's regulatory capital so it can lend again.

Here's the catch. SRTs were originally designed for portfolios of dozens of smaller loans, where risk is naturally diversified. The deals being explored now are different: a few enormous data centre loans, often pointing back to the same handful of tenants. Frank Benhamou, a portfolio manager at Cheyne Capital, told the FT that 'you expect to be paid a bit more for it' when the underlying pool is so concentrated. Investors are also asking that banks keep 'a little bit of skin in the game,' according to David Lucking, a lawyer at Linklaters — meaning the bank retains some exposure so its incentives stay aligned with the buyers'.

A further complication is public opinion. Communities near proposed data centres are increasingly objecting to the noise, water use and electricity demand of the facilities, and that opposition introduces a risk that even the most sophisticated financial structures struggle to price. Moniot suggested that if he were a chief risk officer at a bank and his colleagues kept asking for fresh exposure to multibillion-dollar projects, he'd want to know how easy it would be to 'sell it down' if the public mood shifted.

The wider question this story raises is who actually finances the AI revolution. Banks have hit a ceiling. Private credit firms — Apollo, Blue Owl, Pimco, BlackRock — are stepping into the gap, often with less regulatory oversight than the banks they're partly replacing. The infrastructure of artificial intelligence may end up being owned, in effect, by the same investors who manage retirement portfolios and insurance reserves. Whether that's a clever rerouting of capital or a hidden new pocket of risk is the question Wall Street is now trying to answer in real time.

Source: https://youtu.be/Si9iekby-OM

📎 Download Original ⬇ Download Analysis PDF

📖 Explanation

The biggest banks on Earth are quietly running out of room on their balance sheets — not because of a crash, but because building artificial intelligence is devouring capital faster than they can recycle it.

📖 What's Going On?

JPMorgan Chase, Morgan Stanley, SMBC and MUFG have lent so much money to data centre projects that they're bumping up against internal risk limits. To keep lending, they're now hunting for outside investors — private credit funds, insurers, pension funds — willing to buy slices of those loans off their books.

The trigger is the AI boom. Companies like Oracle and CoreWeave are borrowing hundreds of billions to build the warehouses full of Nvidia chips that train and run models like ChatGPT. One Oracle-linked package alone tops $38 billion across Texas and Wisconsin sites — the kind of size that, in the words of one credit executive, makes banks 'start choking.'

🎯 How To Think About It

Imagine a bank's balance sheet as the cargo hold of a container ship. Every loan is a container; regulators cap how much weight you can carry, and how many containers can be addressed to the same recipient.

💡 Key Things To Know

🌟 Why It Matters

Every chatbot, image generator and AI tutor you use sits on top of physical infrastructure that someone had to finance. If the banks who fund that infrastructure run out of capacity, the whole AI build-out slows — affecting which companies dominate the next decade, where data centres get built (often near your hometown's power grid), and whether public backlash over noise, water and electricity use forces projects to relocate. The article flags growing local opposition as a real risk multiplier.

🔮 The Bigger Picture

We've seen this movie before — in 2007, banks securitised mortgages to keep lending; in the 1980s, they syndicated Latin American sovereign debt. Each time, moving risk off bank balance sheets unlocked enormous growth, and each time it created opaque pockets of risk somewhere else. Watch for two second-order effects: private credit funds becoming the de facto financiers of AI (with less regulation than banks), and rising public backlash over the data centres themselves, which could make these loans far riskier than the spreadsheets currently assume.

📚 Key Terms Glossary

Significant Risk Transfer (SRT)
A financial structure where a bank keeps loans on its books but sells the *credit risk* — the chance of default losses — to outside investors in return for a fee. This frees up regulatory capital so the bank can lend more.
Syndication
The process of splitting a single large loan among multiple lenders so no one bank carries the whole exposure. When syndication 'breaks down,' it means buyers won't take the slices.
Counterparty
The other party in a financial deal. Banks have internal limits on how much exposure they can have to any one counterparty (like Oracle) to avoid catastrophic losses if that party fails.
Private credit
Loans made by non-bank investors — funds run by firms like Apollo, Blue Owl or Pimco — that have grown into a multi-trillion-dollar shadow lending industry over the past decade.
Capital requirements
Regulatory rules that force banks to hold a cushion of equity against the loans they make. Riskier loans require more capital, which limits how much a bank can lend.
Hyperscaler / data centre tenant
A large cloud-computing customer (Oracle, Microsoft, Meta) that leases space inside a data centre and runs AI workloads on it. Lenders treat the tenant's creditworthiness as central to whether the building's loan gets repaid.
Skin in the game
Industry shorthand for retaining some financial exposure to a deal you've sold off, so your incentives stay aligned with the buyers. Investors want banks to keep some, to prove the loans are sound.

✏️ Reading Comprehension Quiz

Tip: log in or create a free account to save your score, earn badges, and appear on the leaderboard. Otherwise the quiz works fine without an account.
Question 1
The passage most directly argues that:
Question 2
Which choice best states the central idea of the passage?
Question 3
According to the passage, banks are pursuing private deals partly because:
Question 4
As used in the passage, the word 'choking' most nearly means:
Question 5
As used in the passage, the phrase 'skin in the game' most nearly means:
Question 6
Which statement about Significant Risk Transfers can most reasonably be inferred from the passage?
Question 7
The passage suggests that growing public opposition to data centres could:
Question 8
The author's tone in describing the banks' situation is best described as:
Question 9
Which choice can most reasonably be inferred about the relationship between banks and non-bank investors in this market?
Question 10
Which line from the passage best supports the answer to the previous question?
← Previous (older)
Beijing Just Killed a $2bn Meta Deal — and Sent Silicon Valley a Message
Next (newer) →
Why the Pentagon Just Froze 165 American Wind Farms