In a recent episode of the DePINed Podcast, host Tom Trowbridge, Co-Founder & CEO of Fluence, sat down with Conor Moore, Co-Founder of USD.ai, to unpack one of the least discussed — yet most critical — layers of the AI boom: credit. While most conversations around AI infrastructure focus on model performance, chips, or hyperscaler demand, this episode turns to a more structural question:
Who finances the GPUs powering AI and why doesn’t traditional finance do it well?
USD.ai is building a crypto-native lending protocol designed specifically for GPU infrastructure. Instead of repackaging traditional private credit products on-chain, the team is attempting to redesign how asset-backed lending works for one of the fastest-growing hardware markets in the world.
Below is a breakdown of the core themes from the conversation.
The Missing Credit Layer Behind AI
AI runs on GPUs. But GPUs run on debt.
NVIDIA alone is expected to sell hundreds of billions of dollars in GPU hardware in the coming year. Over a multi-year horizon, industry estimates project trillions in AI infrastructure spend. Yet behind this explosive growth lies a fundamental bottleneck: structured credit.
GPU clusters require large upfront capital expenditure. Neo-cloud operators and compute providers need financing to deploy hardware before revenue fully materializes. In most industries, this gap is filled by banks, private credit funds, and securitization markets.
In AI infrastructure, that credit layer is still underdeveloped. This is the gap USD.ai is targeting.
Why Traditional Finance Can’t Service GPU Loans
At first glance, GPUs appear to be ideal lending collateral. They are high-value, in constant demand, and generate recurring rental income. But structurally, traditional finance struggles with this asset class.
First, GPUs depreciate quickly — typically over three to five years. Traditional asset-backed securities (ABS) markets rely on aggregating loans and securitizing them over time. That process can take years. By the time a lender originates enough loans to structure an ABS product, the assets may already be close to maturity.
Second, private credit funds tend to focus on large ticket sizes. There is a “missing middle” in the GPU market: loan sizes ranging from a few million to $100 million that are too small for large funds but too complex for banks.
The result is a market where significant demand for capital exists, yet traditional structures can’t move fast enough to meet it.
Turning GPUs into Tradable On-Chain Credit
USD.ai approaches the problem by removing the securitization lag entirely.
Instead of originating loans, bundling them, and waiting to sell them into traditional markets, the protocol executes loans on-chain from day one.
The structure works as follows:
- GPU hardware title is tokenized, creating an on-chain representation of the physical asset.
- Loans are executed on-chain.
- Individual loans are aggregated into a yield-bearing token.
- That token is immediately tradable in secondary markets.
In this model, crypto isn’t an overlay — it is the execution infrastructure.
Investors can mint exposure instantly. Loans become liquid from inception. Depositors gain diversified exposure across borrowers, durations, and jurisdictions without waiting for traditional securitization cycles.
Standardizing Risk: Loan Structure and Insurance
Rather than operating like a bespoke investment bank, USD.ai aims to standardize terms. Typical parameters include:
- ~3-year duration
- 70–80% loan-to-value (LTV)
- Interest rates ranging from roughly 7% for investment-grade offtake to 15% for on-demand compute markets
Rates are structured to balance two sides of the equation:
- Depositors seeking competitive yield (~10% range)
- Borrowers needing capital at rates that compete with leasing alternatives (often ~20% effective cost)
One of the major unlocks for the protocol was adding value insurance coverage from Munich Re. Instead of relying solely on first-loss tranches from private credit providers, the protocol pays a defined premium for full coverage in case of default.
This shift significantly reduces uncertainty. Markets often discount crypto-native credit products due to perceived opacity or default risk. By introducing third-party insurance coverage, USD.ai transforms uncertain yield into more measurable, risk-adjusted return.
Lending to the Asset, Not the Company
A key design decision is that USD.ai lends against the GPU asset itself.
Borrowers establish bankruptcy-remote SPVs (special purpose vehicles) that hold the hardware title and contractual relationships, including colocation agreements and rental contracts. The loan is secured directly against those physical assets.
This non-recourse structure offers several advantages:
- No need for complex corporate credit underwriting
- Faster approval cycles
- Cleaner collateral recovery
In a commoditized compute market, the thesis is simple: the physical asset and its revenue-generating capacity matter more than the broader corporate balance sheet.
Why This Model Isn’t Generalizable
USD.ai did not start here. Early iterations explored generalized NFT-based lending, attempting to build infrastructure for a wide range of tokenized assets. The lesson was clear: real-world credit is not generic.
Each asset class requires deep operational understanding. Market structure, depreciation dynamics, resale liquidity, telemetry visibility — all differ dramatically across industries.
GPUs work because:
- They are high-value assets
- They generate measurable digital performance data
- They operate in a rapidly expanding TAM
- They sit inside one of the fastest-growing industries globally
Rather than building generalized real-world asset infrastructure, the team chose laser focus on GPU infrastructure — a market large enough to support billions in annual lending growth.
Liquidity as a Signal to Capital Markets
Within months, USD.ai accumulated hundreds of millions in deposits and generated billions in secondary trading volume. Liquidity becomes a signal.
If loan terms are fully transparent, publicly visible, and actively traded, they create price discovery infrastructure for GPU credit. Borrowers see where capital is available. Investors observe real demand.
Reaching critical mass — potentially around $1 billion in loans — could act as a forcing mechanism for broader capital markets participation.At scale, the protocol becomes difficult to ignore.
Growth as the Ultimate Risk Protection
Traditional credit markets often obsess over downside scenarios and edge-case structuring.
But in high-growth industries, growth itself provides protection.
Lending to stagnating sectors can carry structural default risk regardless of how complex the capital stack is. Lending into AI infrastructure — arguably the fastest-growing industry globally — changes the equation.
If demand for compute continues expanding at its current pace, borrowers are operating in a structurally growing market.
In that context, GPUs are are exposure to structural technological growth.
Governance, Revenue, and Long-Term Sustainability
USD.ai plans to launch a governance token that controls high-level parameters such as LTV brackets, interest rate buckets, and other structural settings.
Revenue flows primarily from net interest margin and origination fees. Governance determines how that revenue is allocated — whether reinvested into ecosystem growth, distributed, or used for buybacks.
Two distinct risks exist:
- Execution risk — Can the protocol consistently originate and manage quality loans at scale?
- Value capture risk — Is revenue clearly and credibly linked to token economics?
The long-term ambition is not hype-driven exponential spikes, but steady, linear growth — adding billions in loans over time and building durable infrastructure beneath AI and DePIN.
Conclusion
AI runs on GPUs. GPUs run on debt. USD.ai is betting that crypto-native credit markets can finance that debt more efficiently than traditional finance and in doing so, create a new capital layer beneath decentralized infrastructure networks.
If successful, this isn’t just another DeFi protocol. It is the emergence of a new capital market for AI infrastructure.
About DePINed Podcast
DePINed is a podcast exploring the frontier of decentralized physical infrastructure, hosted by Tom Trowbridge, co-founder of Fluence. Each episode features in-depth conversations with founders, builders, and investors who are shaping the future of real-world Web3 networks.
️ Want to be a guest? Fill out the form here
Catch 50+ episodes featuring top DePIN founders and trailblazing protocols on:
