USD.AI explodes to $62.7M in TVL: the “GPU-based” stablecoin drives non-dilutive loans for AI

The Permian Labs protocol has reached $62.7 million in TVL after a Series A round of $13 million led by Framework Ventures, with strategic support from YZi Labs.

In this context, the project aims to provide credit secured by GPU hardware and computing servers, integrating DeFi tools to optimize returns. Official sources from the project and coverage from Permian Labs confirm the details of the fundraising and the initial on-chain metrics.

According to the data collected by Permian Labs and reported by CoinDesk on August 13, 2025, the TVL of the protocol has exceeded $62.7M. On-chain monitoring indicates that active loans remain around 2% of the TVL (approximately $1.25M out of $62.7M).

Industry analysts observe that, in the early stages of RWA (real-world assets) protocols, the conversion from deposits to loans tends to remain low until evaluation and market-making processes are consolidated.

In brief: main numbers and facts

• TVL: $62.7M (on-chain, updated as of August 13, 2025, showing strong growth compared to the initial deposits of the year).
• Round Serie A: $13M led by https://framework.ventures, with participation from Dragonfly, Digital Currency Group, Delphi, and Fintech Collective.
• Support: YZi Labs as a strategic partner for institutional access.
• Active loans: approximately $1.25M, which is ~2% of the total TVL.
• Model: stablecoin pegged to the dollar and backed by GPU hardware and computing servers.
• Yield: integrations with https://www.pendle.finance and systems like AutoVaults (collaborations with K3 Capital, Concrete, and Euler) to optimize returns.

What is USD.AI and who funds it

USD.AI, developed by https://www.permianlabs.xyz, transforms physical computing assets – like servers and GPU – into on-chain collateral, allowing the issuance of a dollar-pegged stablecoin and access to non-dilutive loans for AI operators.

An interesting aspect is the combination of real infrastructure and programmable finance, designed to unlock liquidity without equity surrender. The project closed a Series A of $13 million led by Framework Ventures, with participation from Dragonfly, Digital Currency Group, Delphi, and Fintech Collective. On the institutional front, YZi Labs supports the protocol as a strategic partner.

“USD.AI raises $13M to expand GPU-backed stablecoin lending” — CoinDesk

TVL at $62.7M: timeline and key metrics

Since the private beta launched in June 2025, deposits have grown rapidly, while the demand for loans is proceeding more gradually. It should be noted that the adoption curve of credit tends to consolidate more slowly compared to the deposit phases.

• June 2025: over $10M in deposits.
• July 2025: balance in a range between $30–40M.
• August 2025: peak with over $62.7M of TVL and active loans for about $1.25M.

Compared to the ~$10M initial in June 2025, the TVL has increased by approximately 527%, a figure that reflects strong interest from depositors.

The figures indicate a predominant use of the platform as a yield tool, while the adoption of credit remains in its initial phase. In this context, the conversion from TVL to loans might require more product iteration cycles and greater confidence in risk parameters.

How a hardware-backed stablecoin works

The protocol tokenizes the value of servers and GPU, transforming these assets into on-chain collateral. In exchange, operators can obtain liquidity in stablecoin, retaining corporate ownership of their assets.

Periodic evaluations and automatic liquidation mechanisms are designed to protect the solvency of the system in case the value of the hardware falls below predetermined thresholds. In practice, the risk of price and obsolescence is managed with haircuts and LTV thresholds calibrated on the different classes of machines.

Loans: disbursement times and operations

AI teams can obtain funding in less than a week, unlike traditional channels that require 60 to 90 days. The collateral remains operational, allowing the machines to generate revenue (for example, from training or inference) while securing the loan. It is a dynamic that, if effective, reduces financial friction during growth phases.

Yield and DeFi integrations: AutoVaults and Pendle

For depositors, USD.AI integrates with automatic yield strategies. The AutoVaults, developed in collaboration with K3 Capital, Concrete, Euler, and https://www.pendle.finance, allocate liquidity on instruments that separate and optimize the yield.

Pendle has introduced dedicated vaults connected to the USD.AI ecosystem, incentivizing liquidity and improving market efficiency. In this context, the yield is modulated by incentives and market conditions, so it is not static.

AI hardware credit and non-dilutive loans: the model

The core of the model is credit secured by compute: those who own GPUs and servers can pledge them as collateral, obtain liquidity, and finance growth without having to give up equity.

The solution is designed for teams with computing infrastructures that have limited access to traditional debt. An interesting aspect is the ability to reuse operational cash flows to service the debt, maintaining ownership.

Governance, parameters, and risks

The sustainability of the model depends on strict collateralization parameters and transparent governance. It must be said that the quality of the evaluations remains crucial to align risk and capital usage.

• Key parameters: LTV ratio specific to GPU category, haircut for obsolescence, liquidation thresholds, and stability fees.
• Valuations: periodic updates of the hardware value based on appraisals and market price lists, together with audits on inventories and machine numbering.
• Governance: on-chain management with proposals and voting concerning risk parameters, whitelist for evaluation providers, and emergency policies for liquidations.

Main risks:

• Technological obsolescence: the rapid depreciation of certain GPUs.
• Market: the volatility of DeFi yields associated with deposit strategies.
• Liquidity: a loan growth rate lower than that of the TVL, with the risk of mismatch in scenarios of massive withdrawals.
• Operational: issues related to custody, uptime, and the verifiability of the machine park used as collateral.

Typical use case

An AI operator with a cluster of GPU engaged in training can use them as collateral to obtain liquidity in stablecoin, which can be used for the purchase of additional computational infrastructure or to cover energy costs, while maintaining full ownership of the company. The liquidation mechanism intervenes if the value of the collateral, net of haircuts, falls below the threshold set by the LTV ratio. In this scenario, the risk is managed in an automated way but is not null.

Impact on the ecosystem and next steps

The demand for AI infrastructure financing is rapidly expanding. Solutions like USD.AI channel capital towards physical computing assets, creating a bridge between on-chain finance and real infrastructure.

If the evaluation and risk management criteria consolidate, the model could attract further institutional capital and inspire similar protocols. It remains to be seen how active loans evolve compared to the TVL, a variable that will serve as a litmus test for the economic utility of the system.