Imagine walking into a bank and asking to borrow $22 billion. Not for a house, not for a business you already run — but to buy computer chips you will rent out to other companies. Now imagine ten of the world’s biggest banks saying yes, together, in a single deal. That is what happened this week, per Reuters reporting on September 16 citing Bloomberg News: a group of ten banks — including Goldman Sachs, Sumitomo Mitsui Banking Corp, Barclays, BNP Paribas, and the Bank of Nova Scotia — is providing a $22 billion chip loan to support the Crux AI cloud venture that Blackstone and Alphabet announced back in May.
Let me make sure we have the facts straight, because the structure of this deal is the story. The debt will be used to buy Google’s custom Tensor Processing Units — the specialized chips Google designs for AI work — and the loan is backed by the value of the chips themselves and by customer contracts. Blackstone, for its part, is investing an initial $5 billion in equity to support 500 megawatts of data center capacity expected to come online in 2027. None of the named parties immediately responded to Reuters’ requests for comment. So we have $22 billion in bank debt plus $5 billion in private equity, assembled to build half a gigawatt of AI computing capacity. That is the scale at which the AI buildout now operates: not millions, not billions, but tens of billions per project, raised in a single financing.
To understand why banks are willing to lend at this scale, it helps to see this loan as part of a much larger river. Per a Fortune/StartupFortune piece published September 17, the five biggest AI hyperscalers issued $121 billion in U.S. corporate bonds in 2025 alone — more than four times the roughly $28 billion per year they averaged from 2020 through 2024. Morgan Stanley estimates that AI-related global debt issuance reached about $236 billion by the end of May 2026 and projects roughly $570 billion for the full year. The individual deals are staggering on their own: Meta’s $30 billion bond sale in October, Alphabet’s $25 billion in November, Oracle’s $61.5 billion raised since 2022. The $22 billion Crux loan is not an outlier. It is the new normal — one more large log on a bonfire of AI infrastructure debt that is now measured in the hundreds of billions.
And the banks are not lending blindly. The structure of the Crux deal — debt secured by the chips and by customer contracts — tells you exactly what the lenders believe. They believe the chips hold their value, and they believe the customers will keep paying. This is asset-backed lending applied to the AI age: the collateral is not a building or a fleet of trucks, but silicon and signed contracts for computing power. It is innovative, and it is also a concentrated bet. If demand for AI computing stays strong, the loans perform beautifully. If demand softens — if the AI revenue that is supposed to fill those data centers arrives slower than expected — then ten major banks are holding paper backed by chips whose resale value and contracts whose durability have never been tested in a downturn.
This is where the story leaves the tech pages and walks into every family’s kitchen. Because banks do not have an infinite supply of money to lend. When ten global banks commit $22 billion to AI chips, that is $22 billion of lending capacity directed toward one corner of the economy. And the scale is so large now that it is moving prices for everyone. At his press conference on September 16 — the same day the Crux loan was reported — Federal Reserve Chair Kevin Warsh addressed rising long-term bond yields directly. Per Reuters, he said: “The surge in capital expenditures, which I referenced in my remarks, is real, and the so-called hyperscalers are out in the market raising funding, and so the competition for capital is real.” The Fed chair is telling us, in plain language, that AI companies borrowing hundreds of billions is one of the forces pushing up borrowing costs across the economy. The 10-year Treasury yield closed at 5.02% on September 16, per Reuters. Every family with a mortgage, every small business with a credit line, every saver earning interest is living downstream of these decisions.
It’s not enough to just admire the ambition of a $22 billion loan — we must listen to what it reveals about how capital is being allocated, learn who benefits and who pays when the biggest borrowers absorb the market’s lending capacity, and contribute our own clear thinking about what kind of economy we are building.
So let us think carefully about what this debt means for the economy beyond tech. There are three layers worth separating. The first is the most direct: the AI buildout is real economic activity. Data centers get built, construction workers get hired, electricians and engineers find work, local tax bases grow. Five hundred megawatts of data center capacity does not appear by magic; it is poured concrete, pulled cable, and thousands of paychecks. For communities that host these facilities, the investment is tangible and welcome. Debt, in this sense, is doing what debt is supposed to do: pulling future productive capacity into the present.
The second layer is the financial system’s exposure. When banks lend tens of billions against AI chips and AI contracts, they are making a collective judgment that the AI revenue story will continue. Banks are usually careful, diversified lenders; the fact that ten of them joined this single syndicate suggests genuine conviction — but also genuine concentration. The history of finance is full of moments when careful lenders all reached the same careful conclusion at the same time, about mortgages, about telecom fiber, about any asset class that looked like a one-way bet. I am not predicting that outcome here. I am simply noting that $22 billion from ten banks, plus $121 billion in hyperscaler bonds in a single year, plus a projected $570 billion in AI-related debt issuance, is the kind of concentration that deserves clear-eyed attention rather than celebration alone.
The third layer is the one that touches ordinary savers and borrowers most directly: the competition for capital. Every dollar lent to an AI data center is a dollar not lent to something else, and when demand for capital surges, the price of capital rises. This is not theory; it is the mechanism Warsh described. A family refinancing a home, a young couple buying their first house, a small manufacturer financing new equipment — all of them borrow in a market where AI giants are now among the largest customers. The 30-year mortgage rate climbed to nearly 7% last week, its highest in nearly a year per the Mortgage Bankers Association, as reported by the Journal on September 16. Many forces drive mortgage rates, and AI borrowing is only one of them — but it is a real one, named by the Fed chair himself.
My take: I find myself of two minds about the $22 billion Crux loan, and I think both minds are honest. On one hand, this is capitalism working as designed: banks assessing collateral, pricing risk, and funding productive investment that creates jobs and capacity. The structure — secured by chips and contracts, paired with $5 billion of Blackstone equity taking the first loss — is thoughtful, not reckless. On the other hand, the sheer velocity of AI debt accumulation gives me pause. When Morgan Stanley projects roughly $570 billion in AI-related debt issuance for a single year, we are no longer talking about a sector raising money; we are talking about a sector reshaping the credit markets. My view for ordinary investors is practical: pay attention to the lenders, not just the borrowers. The banks earning fees and interest on this buildout — the Goldmans, the Barclays, the SMBCs of the world — are participating in the AI boom with a creditor’s margin of safety rather than an equity holder’s binary risk. And for families, the takeaway is simpler: in a world where the biggest borrowers are competing for capital, the value of being a saver with options — of carrying less debt, of keeping financial flexibility — only grows. The hopeful note is that none of this is hidden. The deals are reported, the yields are published, the Fed chair is explaining the mechanics out loud. We have the information. What remains is to use it wisely — to listen, to learn, and to contribute our own steady judgment to an economy that is being rewired, one $22 billion loan at a time.
