Bullish is putting $100 million of stablecoin funding behind USD.AI, a protocol that lends against GPUs used in AI infrastructure. The headline is simple enough. The implication is messier: crypto is trying to become a lender for physical compute, not just a casino for tokens.
- $100 million stablecoin debt facility for GPU-backed loans
- Non-recourse financing secured by hardware, not company balance sheets
- Bullish plans to list sUSDai and support trading liquidity
- The deal pushes crypto deeper into real-world asset financing
- The hard part: GPU depreciation, collateral quality, and liquidation risk
According to the companies, Bullish agreed on Friday to provide USD.AI with a $100 million stablecoin debt facility to fund loans backed by graphics processing units, or GPUs. In plain English, Bullish is supplying credit in stablecoins so USD.AI can finance AI infrastructure operators that buy expensive compute hardware.
That is the pitch. The reality is more complicated, because hardware lending only works when the collateral keeps enough value long enough to cover the loan. GPUs are productive assets, but they also depreciate fast. In other words, this is not your grandfather’s mortgage.
USD.AI, developed by Permian Labs, uses non-recourse loans secured by the computing hardware purchased with the financing. Non-recourse means the lender’s claim is limited to the pledged collateral. If the borrower defaults, the lender can go after the GPUs, but not the borrower’s other assets.
That structure can be attractive for operators who want capital without giving up equity or putting their broader business on the hook. It also shifts more risk onto the lender, because the lender is relying heavily on the resale value and utility of specialized hardware that can lose value quickly when newer models hit the market.
And that’s the catch. A GPU that is hot property today can become yesterday’s silicon once the next generation lands. If collateral prices fall hard enough, the lender may end up holding a pile of expensive metal with a lot less punch than expected. No amount of onchain branding fixes that.
Bullish said it used USD.AI’s onchain records when assessing the facility, according to Thomas Cowan, Bullish’s head of tokenization. He said:
“Our commitment to USD.AI reflects a conviction we’ve believed since our first investment in the protocol: that credible, well-structured real-world assets belong onchain.”
There’s a real idea underneath the corporate polish. If a blockchain can help track collateral, loan activity, and repayment flows more transparently, then it can do something more useful than minting another forgettable ticker. But transparency does not erase bad underwriting, and onchain records do not stop a GPU from aging like milk in the sun.
Bullish had already invested in USD.AI before agreeing to this financing. Bullish Capital also invested $4 million in September 2025, so this is clearly not a casual one-off. The exchange and its investment arm are leaning into the same thesis: AI compute can be financed as a distinct credit market.
That’s also how David Choi, CEO of Permian Labs, framed it:
“Compute is becoming a credit market in its own right, ”
He’s not wrong. AI infrastructure is expensive, and the operators building it need financing. The market has already proven there’s demand for capital tied to compute, and that demand can be served in ways that don’t involve handing over equity to every new fund with a pitch deck and a caffeine problem.
According to the announcement, USD.AI will use the $100 million facility to originate loans for middle-market AI infrastructure operators, mid-sized companies building or running AI compute infrastructure, smaller than hyperscalers but still hungry for capital. The loans are designed to be non-dilutive, which means the businesses can borrow without giving up ownership.
USD.AI has already shown the model can attract serious capital. In June, it announced a $98.1 million loan backed by 2, 304 Nvidia B300 GPUs. Investors also fully funded another $34 million facility secured by 768 Nvidia B200 units. Combined, those two disclosed loans totaled 3, 072 GPUs and more than $132 million in financing.
That doesn’t prove the model is bulletproof. It does show this is no longer a tiny side experiment. It’s a live credit market built around hardware that powers AI workloads, with enough scale to attract larger financial players.
Bullish also plans to list sUSDai, USD.AI’s yield-bearing token, across several trading pairs on its institutional exchange. The company will also support the token with a dedicated market-making program once trading begins.
For readers who don’t live and breathe market structure, market making is what helps a token trade smoothly. Market makers place buy and sell orders to add liquidity and narrow spreads. That can make it easier for users to enter and exit positions, but it also deserves scrutiny if the support is thin or opaque. Liquidity is useful. Fake depth is just theater with spreadsheets.
Bullish and USD.AI did not disclose the planned trading pairs, launch date, or market-making budget. Those missing details matter. Without them, it’s hard to judge how serious the rollout is or how much support will actually exist once sUSDai hits the market and traders start doing what traders do.
In broader terms, this is another sign that crypto is moving deeper into financing physical infrastructure. That includes real-world assets, tokenized claims, and onchain credit tied to tangible hardware. The upside is obvious: faster capital formation, more flexible lending, and potentially better transparency than traditional closed-door finance.
The downside is just as obvious. Specialized hardware depreciates. Secondary markets can get thin. And if underwriting gets loose, “real-world assets” can quickly become “real-world headaches.” The blockchain doesn’t abolish risk. It just gives risk a nicer interface.
Bullish’s move also fits the company’s effort to position itself as more than a trading venue. It operates under the European Union’s Markets in Crypto-Assets framework as an authorized crypto asset service provider offering spot trading and custody, and it secured a New York BitLicense in September 2025. Those are the kinds of regulatory credentials that help a firm support listings and liquidity programs without looking like it was assembled in a garage between two espresso machines.
The public-market backdrop is messy, too. Bullish completed its New York Stock Exchange debut in August 2025, pricing shares at $37 and raising about $1.03 billion. The stock opened at $90 in its first session, implying a valuation of about $5.4 billion. On Friday, shares were trading around $33, according to Yahoo Finance, leaving BLSH more than 60% below that opening price.
Still, Bullish has picked up some momentum. Shares have gained about 45% over the past month. Other U.S.-traded crypto names have also moved higher over the same period, including Strive, which gained about 88%, Canaan, which rose roughly 55%, and Circle, which added close to 40%. That kind of action shows investors are still chasing crypto exposure, but it also reminds everyone that enthusiasm in public markets can disappear fast.
The compute-finance theme is spreading beyond Bullish and USD.AI. The source notes that Injective introduced an Nvidia GPU derivatives market in August 2025, and Aethir and Injective launched a tokenized GPU marketplace in December 2024. Different products, same underlying idea: if GPUs are becoming economically important assets, financial rails will follow.
That doesn’t mean every GPU-linked crypto product is destined to matter. Some of this will be durable infrastructure. Some of it will be a dressed-up wrapper around speculative yield. The difference will come down to underwriting, collateral management, liquidation mechanics, and whether the economics still work when the market stops applauding.
For now, Bullish’s $100 million commitment to USD.AI is a meaningful bet on a simple proposition: AI compute can be financed like a real asset class. If the collateral holds up and the credit stays disciplined, that could be useful. If not, it’s just another clever structure waiting to find out what depreciation really feels like.
Key questions and takeaways
-
What did Bullish agree to do?
Bullish agreed to provide USD.AI with a $100 million stablecoin debt facility to support loans backed by GPUs used in AI infrastructure. -
What makes the loans different from normal business lending?
They are non-recourse, which means the lender can claim the pledged hardware if there’s a default, but not the borrower’s other assets. That can help operators, but it also makes collateral quality critical. -
Why are GPUs a risky form of collateral?
GPUs can lose value quickly as newer models arrive. If resale prices fall too far, the lender may not recover the full amount of the loan. -
What is sUSDai?
sUSDai is USD.AI’s yield-bearing token. Bullish plans to list it and support trading with market-making, which should improve liquidity if the rollout is well executed. -
Does this make onchain credit more credible?
It helps, but only if the underwriting is solid. Blockchain records can improve transparency, yet they do not remove the basic risks of falling hardware values, weak collateral, or sloppy loan terms.
The big takeaway is simple: crypto is getting more useful when it helps finance actual productive assets instead of just recycling leverage around itself. The ugly truth is that the same tools can also be used to package risk into something shiny and tradable. As always, the collateral is where the dream either holds up or gets liquidated.
Further reading
A few related pieces for the bigger picture on tokenized infrastructure, rails, and AI-linked stablecoin plays.
