US and UK regulators are giving the biggest banks more room to lend just as AI infrastructure is becoming one of the most capital-hungry markets in the world.
The biggest banks have spent years watching private credit funds take territory they used to control. Now the regulatory tide is moving in their favor, and the timing is hard to ignore. A looser capital regime in the US and UK could give top lenders a combined $1.3 trillion of additional deployment capacity, according to the Financial Times, opening a new contest over who finances the next phase of the AI buildout.
This is not just a banking story. It reaches directly into technology markets because AI is no longer funded only by venture rounds, cloud budgets and chip prepayments. Data centers, power contracts, fiber networks and GPU clusters require debt on a scale that looks more like energy infrastructure than software. When the rules around bank balance sheets change, the cost and availability of that money changes too.
In the US, the shift has been building for months. Reuters reported in April that big American banks could release up to $320 billion in capital under revised draft rules, citing Morgan Stanley analysts. The Federal Reserve has said softened Basel and global systemically important bank surcharge proposals could reduce capital levels at large banks by between 4.8 percent and 7.8 percent. The Congressional Research Service later described the broader package as a net reduction in required Tier 1 capital of 5.6 percent to 7.9 percent, depending on the bank category.
That is dry regulatory language, but the market effect is simple enough. Capital is expensive. If a bank has to hold less of it against certain assets, it can do more lending, take more underwriting risk, return more cash to shareholders, or some combination of all three. The practical question is where that freed-up capacity goes first.
The obvious destination is AI infrastructure. Banks including JPMorgan Chase, Morgan Stanley and Japan's SMBC have already been trying to manage their exposure to large data center debt packages. Silicon UK, citing earlier Financial Times reporting, noted this month that lenders have explored selling loan stakes and using risk transfer structures after reaching limits on AI data center financing.
That tells you something important. The banks are not stepping away from AI infrastructure because they dislike the opportunity. They are trying to make room for more of it. A $38 billion construction debt package tied to Oracle-leased data centers in Texas and Wisconsin has already shown how large these deals can become. Traditional corporate lending desks were not built for a world where one compute project can look like a megaproject.
The Federal Reserve's May financial stability report showed why supervisors are watching carefully. Market contacts surveyed by the New York Fed cited AI-related risks, including equity valuations and debt-funded capital spending. They also pointed to private credit as an increasing concern, with pressure from redemptions and weaker sentiment potentially spilling into broader markets.
That creates an awkward policy balance. Regulators want banks to support growth and compete globally, but they do not want the post-2008 safety framework to be weakened just as leverage is returning through a new technology boom. The strongest argument for easing is that banks are supervised, stress-tested and transparent compared with much of private credit. The strongest argument against it is that memories of the last credit cycle tend to fade just when capital feels most abundant.
Private credit finally has a serious challenger
For the startup ecosystem, the most interesting effect may come downstream. When large banks have more room to underwrite big corporate and infrastructure deals, they also become more competitive in venture lending, acquisition finance and bridge facilities. That matters for later-stage startups that need debt to extend runway, finance equipment or support acquisitions without immediately raising more equity.
Private credit funds captured market share when banks became more constrained. They moved faster, negotiated directly and filled gaps in leveraged lending, software financing and asset-backed structures. But they also charge for that flexibility. If banks can now offer cheaper capital while keeping distribution channels into syndicated loan and bond markets, borrowers will have more leverage in negotiations.
This will not make money easy for every AI startup. Banks still prefer contracted revenue, collateral, strong sponsors and visible cash flows. A model company burning cash without durable enterprise demand will not suddenly become a bankable infrastructure borrower. The winners are more likely to be data center developers, power providers, cloud-adjacent businesses and AI companies with hard assets or long-term customer commitments.
Crypto custody is another beneficiary to watch. Large banks have wanted a clearer role in digital asset custody and tokenized markets, but capital and compliance costs have slowed their moves. A friendlier balance-sheet environment could let global banks test those businesses more aggressively, especially where institutional clients want custody, settlement and lending handled by regulated names rather than specialist crypto firms.
The risk is that a capital release becomes a race for volume. Banks will argue they are better placed than shadow lenders to finance strategic infrastructure. Private credit managers will argue they understand the risks and can customize capital more effectively. Both can be right for different borrowers. The market will decide deal by deal.
What matters now is whether this new capacity is used to fund durable infrastructure or simply to stretch another cycle. AI needs enormous investment, and the financial system is moving to meet it. For founders and investors, the signal is clear: bank capital is becoming relevant again, and the next financing conversation may not end with a venture fund or a private credit term sheet.
Also read: Infratil is turning Australasia into an AI data centre bet • AI-made lawsuits are forcing courts to write new rules. • A quantum AI test on IBM hardware points to a new compute race
