Bullish Backs GPU-Collateralised Credit with $100m Stablecoin Facility

Crypto exchange operator Bullish has committed a $100m stablecoin credit facility to USD.AI, a protocol that lends against graphics processing units, extending an existing investor relationship into direct balance-sheet financing for the AI compute buildout.

The facility, announced on 28 August 2026, gives USD.AI stablecoin liquidity to originate non-recourse loans secured against high-performance computing hardware. It is not Bullish's first exposure to the protocol: the NYSE-listed company was already an investor in USD.AI's August 2025 Series A, a $13m round led by Framework Ventures that also drew in Dragonfly, Arbitrum and Binance venture arm YZi Labs. The new commitment moves Bullish from equity backer to lender, and it lands as GPU-backed private credit matures from a niche experiment into one of the larger asset classes forming on-chain.

What is Bullish actually financing?

USD.AI, developed by Permian Labs, runs a dual-token model: USDai, a fully collateralised dollar stablecoin, and sUSDai, a yield-bearing counterpart backed by income-producing compute loans. Borrowers post GPU hardware as collateral, which is tokenised on-chain and verified for ownership and insurance. Loans are non-recourse and secured solely by the underlying kit, isolating the risk from an operator's corporate balance sheet.

The economics are built around hardware depreciation. USD.AI loans amortise over three years with monthly principal and interest payments structured so that roughly half the original loan is repaid within the first year, pulling the loan-to-value ratio down toward 20 to 30 per cent by year two. That amortisation profile is the protocol's answer to the central risk in compute lending: that a GPU loses market value faster than the debt against it is retired.

The protocol has scaled quickly. Its mid-2026 reporting put total value locked at close to $400m, with around $200m deployed into active loans, and it has already written facilities including a roughly $98m three-year deal to fund the deployment of NVIDIA B300 GPUs and a facility of up to $500m arranged for Australian operator Sharon AI. Against that book, a single $100m commitment from one counterparty is a material addition rather than a marginal top-up.

Why does compute financing need a new credit market?

The pitch from both companies is that AI compute has become a credit market in its own right, and that traditional lenders are poorly equipped to underwrite it. GPUs are expensive, fast-depreciating and in constant demand, which makes them awkward collateral for banks but attractive for asset-backed structures that can value and liquidate hardware quickly. USD.AI has claimed loan approval times materially faster than conventional lenders, a function of tokenising the collateral rather than running it through a traditional credit committee.

The scale claim is worth stating plainly rather than in the promotional register of the original announcement. Bullish frames GPU-backed lending as a segment that has grown to eclipse legacy consumer debt markets such as auto loans and home equity lines of credit. That is a directional argument about where private-credit demand is heading, driven by hyperscaler and neocloud capital expenditure, rather than a settled measure of outstanding balances, and readers underwriting exposure should treat it as such.

What does the sUSDai listing change?

The financing is paired with a market-structure component that may matter more over time than the headline number. Bullish plans to list sUSDai across multiple trading pairs on Bullish Exchange, supported by a dedicated market-making programme. The intent is to build secondary liquidity and price discovery for GPU-backed debt, so that exposure to compute credit can be traded rather than simply held to maturity.

That is the part institutional allocators will watch. A tradable, priced instrument for compute-backed credit begins to answer the liquidity problem that has kept most real-world-asset lending illiquid and hard to mark. Whether the listing generates genuine two-sided depth or thin, sponsor-supported volume is the open question, and it will not be answerable until the markets are live.

What relationships should readers keep in view?

Two connected-party relationships sit behind this announcement and neither is incidental. First, Bullish is not a new arrival underwriting USD.AI at arm's length; it is an existing shareholder increasing its commitment, which shapes how independent the underwriting diligence described in the announcement can be taken to be. Second, Bullish is the parent company of CoinDesk, one of the most widely read outlets covering digital assets, which reported the facility on the day it was announced. That is a legitimate ownership structure, but the overlap between the financier of a deal and a primary source of coverage on it is context readers deserve when weighing the reporting.

Bullish itself carries relevant history here. The company listed on the New York Stock Exchange in August 2025 under the ticker BLSH, raising roughly $1.1bn, and took the unusual step of receiving about $1.15bn of its IPO proceeds in stablecoins. Its European entity is registered as a crypto-asset service provider under the EU's Markets in Crypto-Assets regime. A firm that settled its own listing in stablecoins extending a stablecoin credit line to a lending protocol is consistent strategy, not coincidence.

Why this matters to FinanceX readers

Compute-backed credit is where two capital-intensive stories intersect: the AI infrastructure buildout that needs financing at scale, and the tokenisation of real-world assets that has been searching for a use case with genuine institutional demand. A $100m facility from a listed exchange operator, paired with a plan to make the resulting debt tradable, is a test of whether GPU lending can graduate from crypto-native yield product to an asset class allocators can underwrite, price and exit.

For investors, the signals to track are secondary-market depth in sUSDai once listed, the performance of the amortisation model through a hardware downcycle, and how much of the reported diligence reflects genuinely independent risk assessment given the pre-existing investor relationship. The demand for AI compute is not in doubt; whether the credit market being built around it holds up under stress is the part still to be proven.