GM. This is Milk Road, your crypto cheat code - minus the part where you still FOMO at the top.
Microsoft is spending $80B on AI infrastructure this year.
Google committed $75B.
Amazon is in for $100B.
π These are the companies that can write nine-figure checks for GPU clusters without blinking.
Most AI companies are not Microsoft.
They need compute, but they cannot self-fund it.
And the traditional lenders, banks, and equipment finance firms have no idea how to underwrite a rack of H100s.
π₯ The AI infrastructure boom has a financing problem.
The companies building out compute clusters need capital.
Buying enough NVIDIA H100s to run a serious AI operation costs tens of millions of dollars.
Most operators don't have it sitting around.
That gap is a lending opportunity:
- GPUs are productive assets.
- They generate real cash flow from compute rentals.
Imagine you can lend against that hardware, charge interest, and pass the yield to stablecoin holders.
No delta exposure to crypto markets. No speculation on token prices.
π Just AI compute rental income flowing to you as a depositor.
This is the "InfraFi" thesis, using physical AI infrastructure as the backing for onchain financial products.
It is one of the most genuinely interesting yield ideas in crypto right now.
Think about what it would mean if it worked.
A stablecoin yielding 10 to 15%, backed not by funding rates or emissions but by the physical hardware powering every AI model you use.
Sounds pretty promising, right?
So I looked hard at USD.AI.
The protocol most people point to when discussing this thesis:
- The team has real credentials.
- The backers include Framework and Dragonfly.
- The concept is sound.
But here is the problem: the product does not match the pitch.
Not yet.
Why? Thatβs what we are going to reveal today and more.
Here is what we cover:
- Why GPU-backed yield is different and why it is actually compelling.
- The gap between USD.AI's marketing and reality.
- What role PayPal actually plays in all of this.
- Is their governance token CHIP worth buying once it is live?
USD.AI's token generation event is coming up soon. This is our chance to get ahead of it and know exactly whether to buy or walk away.
The pitch is compelling. But I don't invest in pitches.
I invest in numbers, execution, and evidence that a business model works in practice - not just on paper.
So let's dig in together and see what's actually there.
Who is behind USD.AI
Before we talk about what they are building, letβs look quickly about who is building it.
USD.AI is built by Permian Labs, a Delaware company founded in 2021.
The three co-founders:
- David Choi (CEO, formerly Deutsche Bank investment banking, co-founder of MetaStreet),
- Conor Moore (COO, Deutsche Bank, Rockpoint Group private equity),
- Ivan Sergeev (engineering, FPGA hardware background, DRW).
All fully doxxed. No known controversies and with crypto history.
MetaStreet, their prior project, used the same architecture as USD.AI but for NFT-collateralized lending.
It reached $400M in cumulative loan volume, then collapsed when the NFT market did.
That history cuts both ways:
- β It proves they can build and operate this kind of protocol.
- β It also showed the model's sensitivity to collateral market cycles.
But this time, GPUs are a lot more reassuring as collateral than JPEGs.
π The team is experienced, with credible backgrounds and a track record of building in this space. They spotted a better opportunity and made the pivot.
That part is not the problem.
But letβs talk about the product.
The product
At its core, USD.AI is not doing anything wildly different from yield-bearing stablecoin issuers you might already know, like Sky or Ethena.
The basic idea is the same: deposit dollars, earn yield.
But the source of that yield is where things get interesting.
As usual, the protocol runs on two tokens.
1. USDai is the simple one.
It is a fully-backed synthetic dollar. Deposit one dollar, get one USDai. Redeem one USDai, get one dollar back. No yield, no tricks.
While your capital sits waiting to be deployed into GPU loans, it sits in PYUSD. It is PayPal's stablecoin, earning yield from T-bills and a promotional rate PayPal is running to boost adoption.
2. sUSDai is where the yield lives.
It is built as an ERC-4626 vault, a standard for yield-bearing tokens on Ethereum where the token itself appreciates over time, rather than paying out separate interest.
Think of it like a savings account where your balance quietly grows, rather than a checking account that receives deposits.
Current yield sits between 5.9%. You can exit, but not instantly. Withdrawals go through a 30-day batch queue called the QEV.
On the surface, the pitch holds up. You lock up liquidity, you get paid a juicy yield in return.
A fair trade on paper.
But you need to understand what is actually happening under the hood.
Is the yield guaranteed?
The pitch is simple.
π Deposit dollars, earn 5.9%, backed by real GPUs you could theoretically touch.
That feels a lot more solid than some algorithmic stablecoin held together by vibes.
But the yield is not guaranteed.
It comes from loans to real companies. Real companies go bust.
When they do, someone eats the loss. The question is whether that someone is you.
Here is how the machine works.
AI labs need GPUs badly.
Buying them outright is brutal, so they borrow money and pledge their hardware as collateral. The interest they pay flows back to sUSDai holders. That is the engine.
But this creates a weird problem: how do you use a physical GPU sitting in a data center as collateral on a blockchain?
USD.AI built a three-layer answer.
Layer 1 turns the hardware into an onchain asset. The GPU gets locked in an insured data center.
A system called Caliber then mints a GWRT NFT, basically a digital ownership certificate modeled after real U.S. commercial law. Novel structure. Never tested in court.
Layer 2 is your first cushion. FiLo Curators are third-party originators who take the first hit if a borrower defaults. Think of them as specialist lenders who sit between the borrowers and your money.
Problem is, nobody knows who they are or how much capital they actually have on the line.
Layer 3 is Munich Re, one of the biggest reinsurers in the world, covering the next wave of losses. How much coverage exactly? Also unknown.
Only after both layers are gone does your vault start bleeding.
Below is the full picture, tracing your deposit all the way through to the yield, with every safety layer mapped out.
Here is the full picture in one diagram.
Follow the money from your deposit all the way through to the yield coming back, and see exactly where the safety layers sit.
If the protocol reaches its end state, GPU loans dominate and compute rental yield flowing to sUSDai holders, this is a genuinely differentiated financial product.
Thatβs the product we would love to use.
Getting there is the whole question, because right now the gap is pretty huge.
Uh, Ohβ¦ π§ The rest of this report is exclusive to PRO members!
Already a PRO member? Log in here.
WHATβS LEFT INSIDE? π
- What is their biggest problem right now?
- How much loan volume do they have in the pipeline?
- How do they actually make money?
- Is CHIP worth buying after it launches?
Upgrade your subscription today to unlock access to all of the milky insights above, PLUS:
- Weekly reports to help you manage investments, allocate capital, take profits, and stay ahead in crypto. π
- Access to the PRO Community, where the Milk Road crew & 1000s of fellow PROs talk crypto. Donβt miss the monthly live events! π«
Already a PRO member? Log in here.
