🟪 Crypto can originate-and-distribute now

How I became a data center financier.

“A bank is therefore not an office for ‘borrowing’ and ‘lending’ money, but it is a Manufactory of Credit.”
— Henry Dunning Macleod, The Theory of Credit (1891)

Crypto can originate-and-distribute now

Roughly speaking, there are three ways a bank can make money from deposits.

One: clip coupons.

We accept low or no interest on our bank deposits in return for the convenience of having a checking account to receive a paycheck into and pay bills out of. Banks can make money by investing those deposits in risk-free government debt (or parking them with the Fed). Banks that do only this are called “100% reserve” or “narrow” banks. There are not many modern examples, in part because the Fed discourages it (they want banks making loans).

A notable exception is stablecoin issuers. Circle, for example, is a narrow bank: it has all $77 billion of its deposits (the dollars it’s received in return for USDC) invested in cash and cash equivalents. Tether is an almost-narrow bank: it has 75% of its $189 billion of assets in cash and cash equivalents (the rest is in a mishmash of bitcoin, gold, equities, and loans).

Two: originate and hold.

This is the model that most people think of when they think of banks: taking deposits and making loans.

(Or creating deposits by making loans. It depends how you look at it.)

Traditional banks like the Bailey Brothers Building and Loan would simply hold the loans they make until maturity, earning a spread on the difference between what they pay depositors and what they charge borrowers. This had the benefit of making banks think hard about what loans it should make. It also had the disbenefit of creating duration mismatch — the short-dated liabilities (on-demand deposits) and long-dated assets (mortgages, say) that make banks subject to runs.

As George Bailey told his panicked depositors: “The money’s not here. Your money’s in Joe’s house — that’s right next to yours — and the Kennedy house…”

The closest crypto equivalent to this model might be CeFi lenders like the now-defunct Celsius, which attracted deposits by offering interest rates as high as 18.5% — approximately 200 times what a bank would pay on a checking account.

“Somebody is lying,” Celsius CEO Alex Mashinsky liked to say. “Either the bank is lying or Celsius is lying.”

The banks were not lying: borrowing short and lending long is a tough business.

(And Mashinsky is serving a 12-year sentence for defrauding customers.) 

Three: originate and distribute.

Here, banks make a loan and then sell it to investors, which earns the bank fees and a spread.

More importantly, it frees up their balance sheet to make more loans. Your bank probably doesn’t want your mortgage on its books for 30 years (even at 7.4%), so they likely sold it to Fannie Mae or Freddie Mac, which then pooled it with hundreds of others and sold it to investors as a mortgage-backed security.

As of recently, crypto has that kind of banking, too: the USD.AI protocol takes deposits, originates loans, and sells them.

The protocol’s depositors receive USDai, a stablecoin that USD.AI issues against USDC.

USD.AI then swaps the USDC it receives for US dollars, which it uses to make loans collateralized by GPUs. 

Finally — as of just last week — the loans USD.AI originates can be sold on its new GPU Loan Exchange (GLX).

“Going forward, USD.AI will process transactions through sUSDai initially,” the announcement said (in reference to the staked version of its stablecoin), “and then recycle loans into permanent capital sources through GLX, evolving from a balance sheet lending platform into an exchange.” 

Henry Dunning Macleod might say instead that this evolves them into a Manufactury of Credit, since that’s what they’re doing: USD.AI manufactures loans for sale.

But I’d just call it a bank.

Either way, it sounds pretty impressive because originating and distributing loans onchain against real-world collateral is complicated.

Origination terms, collateral documentation, and disbursements have to be tracked and verified onchain. Financing has to be arranged through SPVs. Real-world collateral has to be contractually owned by bankruptcy-remote entities.

Amazingly, all of this can be arranged faster onchain than it is offchain.

“Smart contract-enforced waterfalls, tokenized collateral, and onchain settlement compress origination timelines from quarters to weeks,” Blockworks analyst Nick Carpinito explains in the definitive note introducing the DeFi sector of collateralized onchain infrastructure lending (COIL).

This is the “core proposition” of COIL protocols like USD.AI, Carpinito adds — and particularly helpful in the market for GPUs, where no one wants to wait quarters for new compute capacity to come online.

Another helpful proposition of onchain lending is the smaller loans it enables. COIL, Carpinito says, “addresses a financing gap that traditional project finance abandons below $50M deal size, where borrowers are locked out of institutional credit markets.”

It does the same for the other side of the equation, too, enabling smaller lenders that are otherwise locked out of institutional markets to create credit.

Like, really small: I just deposited $1,000 into USD.AI — mostly so I can tell people that I’m financing a data center.

And I look forward to financing some robotaxis, too, as soon as the Fractals.finance COIL project is ready to take my money.

None of this is investing advice, of course. If you’re considering becoming an onchain lender, see Carpinito’s note for a looooong list of risks.

But manufacturing credit can be fun — especially when it’s for something useful. 

(And how often can we say that in crypto?)

Brought to you by:

Meridian 2026 is Stellar's annual gathering for the institutions, fintechs, and developers putting financial infrastructure onchain.

Join them October 28-29 at Convento do Beato in Lisbon for two days on tokenization, payments, and what it takes to run them in production.

Register today with Blockworks10 for 10% off.