CHIP, the governance token of the USD.AI protocol, surged 26% on May 8 to $0.068, pushing daily trading volume past $200 million. The token launched on April 21 with simultaneous listings on Binance, Coinbase, and OKX. It hit an all-time high of $0.1383 two days later. Then it lost half its value in two weeks. Now it is bouncing.
The price action is noisy. The protocol underneath is not.
GPU operators borrow against their hardware. That’s the entire pitch.
USD.AI is a lending protocol that lets companies running AI compute infrastructure tokenize their GPUs as collateral and borrow against them. The loans generate yield for depositors through USDai, a synthetic dollar, and sUSDai, its yield-bearing version. CHIP is the governance token: holders vote on risk parameters, interest rates, and which loan facilities get approved.
The use case is specific. Banks do not lend against GPU racks. Nvidia hardware depreciates. Traditional credit markets do not know how to underwrite compute infrastructure. USD.AI is trying to build that credit layer on-chain. The protocol currently holds $344 million in total deposits and has originated $225 million in loans since 2025. The target APR range is 10-15%. Current sUSDai yield: 7.26%.
That puts it at the intersection of three hot narratives: AI infrastructure, real-world asset tokenization, and DeFi lending. Whether the protocol delivers on all three is the open question. But the product is live, the loans are real, and the borrowers are actual GPU operators. That is more than most AI tokens can say.
80% of supply locked. Unlocks start this month.
Here is the part the rally does not price in. CHIP has 10 billion max supply. Only 2 billion are circulating. That is 20%. The remaining 80% is split between core contributors, investors, ecosystem bootstrapping, and reserves. Investor and team tokens have a 12-month cliff followed by linear vesting through 2027-2028.
The first major unlock window opens in May 2026. If new supply hits the market faster than demand absorbs it, price compresses. This is the structural risk that every early-stage token faces. During rallies, nobody talks about it. During drawdowns, it becomes the only conversation.
The volume-to-market-cap ratio reinforces the concern. At launch, CHIP was doing $1.3 billion daily volume against a $200 million market cap. That is 6-7x turnover per day. Even now, $200 million daily volume against a $135 million cap means the token changes hands more than once every 24 hours. That kind of velocity is speculative, not structural.
The AI compute lending thesis is real. The token economics are not proven.
Separate the protocol from the token. The protocol has $225 million in executed loans, $344 million in deposits, and partnerships with enterprise GPU operators. It targets a $1.5 billion pipeline. The first $100 million loan cycle is completing this quarter, which will produce the first real performance data on default rates, yield consistency, and collateral recovery.
The token is different. CHIP is governance-only. It does not capture protocol revenue directly. Holders vote on parameters but do not receive fees. That creates a valuation disconnect: the protocol can succeed while the token underperforms, if governance rights alone do not justify the FDV. At $135 million market cap and $685 million fully diluted, CHIP trades at roughly 3x its total deposits. Whether that is cheap or expensive depends entirely on what the first loan cycle data shows.
The AI infrastructure debate is accelerating across the industry. USD.AI is betting that the hardware powering AI agents needs its own financial layer. That bet has Bitcoin miners pivoting to AI hosting as a tailwind. Whether CHIP is the right way to play it is a separate question from whether the thesis is right.
