Proof of silicon. For every GPU you finance.
A loan against GPUs rests on hardware the lender never sees. Paralon Proof verifies GPU collateral from the servers themselves: every card present, the one that was financed, online and measured, live instead of in a monthly report.
For GPU-backed lenders, onchain protocols and the operators they finance. Pilots are open.
Present
Every GPU behind the loan is identified and checked against the list that was financed. A card missing or swapped is an alert that hour, not a finding at the next audit.
Measured
Every card is made to do work and measured on the card itself, again after every reboot and whenever it is idle. A GPU is known by what it delivers, not by the name it reports.
Sealed
Every report is signed. Once a day, a facility's reports become one proof, published onchain, that a lending contract can read and anyone can check.
The gap
GPU-backed debt is now a market of its own. The collateral is checked when the loan is made, and then mostly through what the borrower reports.
The lender
Holds a lien on GPUs it has never seen run.
What it knows comes from invoices at the start and from the borrower's own reports after.
The protocol
Puts the loan onchain. The hardware stays off it.
A token can say a GPU exists; a contract has no way to check that it still does.
Paralon Proof
Asks the machines, and keeps asking.
The checks that already measure every GPU on Paralon, pointed at the collateral of a loan.
GPU collateral monitoring
One view per loan: every GPU behind it, whether it is there, whether it is the one financed, whether it runs. The reports underneath arrive every minute, one per server.
Example facility
8 servers · 64 × B300
the collateral of one loan, as the lender sees it
- Present64 / 64 GPUs
- Identitymatches the financed list
- Online, last 30 days99.6%
- Measured, not declaredVRAM, bandwidth, compute
- Last measuredtoday, 06:12 UTC
Paralon ProofReports, as they arrive
illustrative- 06:52:00b300-018/8 · 95% busy8.0 kW9727…47ff
- 06:51:09b300-088/8 · 94% busy6.9 kWd671…8bc9
- 06:50:08b300-078/8 · 93% busy7.1 kW44dc…123c
- 06:49:07b300-068/8 · 92% busy7.3 kW0f3e…fdce
- 06:48:06b300-058/8 · 91% busy7.5 kW8b4c…8d18
- 06:47:05b300-048/8 · 90% busy7.7 kWf08a…8c36
- 06:46:04b300-038/8 · idle, measured7.9 kW5c15…6e5d
One report a minute from every server, each signed, each kept.
How a pilot runs
We build the collateral side with the first lenders and protocols, around what their loan terms actually test.
- 01
Bring one facility
The hardware, where it runs, and what your loan terms test.
- 02
We connect it with you
Read-only, and nothing reaches in. Your operations and your customers' work run as before.
- 03
The lender sees it live
Every GPU behind the loan, alerts when the collateral changes, a monthly report to file.
- 04
Onchain, if the loan is
A daily proof on the chain the loan lives on, and a feed a contract can read.
Proof of reserves, for GPUs
Every signed report of a facility goes into one proof a day, published on the chain the loan already lives on. A contract reads the facility's feed; anyone can check any single report against it.
Who it is for
Lenders, protocols and the operators they finance
See the collateral, not a report about it
- Every financed GPU: present, the right one, online and measured
- Alerts when the collateral changes, not at month end
- A monthly report signed by the machines, not written by the borrower
- If the borrower stops paying, the hardware can earn while you decide
An oracle for GPU collateral
- Loans and tokens live onchain; the hardware does not
- A daily proof of the collateral, checkable by anyone
- A per-facility feed a contract can act on
- Published on the chain your loans already use
Prove the capacity you sell
- A public proof page for the GPUs you choose to show
- A badge for your site that links to it
- The same proof your lender asks for, set up once
- Capacity between contracts rents through Paralon if you want
What runs today
Live, and in pilot
Paralon runs on GPUs other people own, so it was built not to take a GPU on its word. How we verify a GPU.
- Every card measured on the card itself: memory, bandwidth, computeLive
- Measured again after every reboot, after each rental, and daily when idleLive
- One machine, one identityLive
- Reports every minute; a server that stops answering known within minutesLive
- Revenue, for capacity rented through ParalonLive
- The same view for GPUs that never rent through ParalonPilot
- Matching every GPU to the financed listPilot
- Signed reportsPilot
- Alerts on the terms of the loan, and a monthly report to filePilot
- A daily proof onchain and a feed per facilityPilot
When collateral goes idle
Idle GPUs lose value every day. These can earn.
If a borrower stops paying, the same servers can rent on the Paralon marketplace where they stand: by the minute, as spot capacity, or reserved for buyers the lender chooses.
Nothing is shipped and nothing is re-racked.
The honest part
What proof does not prove
Not what the hardware is worth
Proof says the GPU is there and works. What it is worth is a valuation, and stays with the firms that do valuations.
Not revenue earned elsewhere
We know the revenue of capacity rented through Paralon. What the borrower earns outside it is shown as declared, next to what the machines show.
Not that a busy GPU is a paid one
Utilisation shows a card at work, not who pays for it. Patterns that look manufactured are flagged; a paying customer is not proven by a busy card.
GPU collateral, answered
How do lenders verify GPU collateral today?
Mostly through documents: purchase invoices, serial lists, the data center's lien waiver and reports the borrower sends, often monthly. What happens to the hardware between two reports is largely unseen. Paralon Proof adds a live view, read from the servers themselves.
What does Paralon Proof verify?
That each GPU behind a loan is present, is the one that was financed, is the card it claims to be by measurement, is online, and stays so, every minute rather than once a month.
Does the borrower have to rent the GPUs through Paralon?
No. The collateral keeps running the borrower's own work or their customers'. Renting spare capacity through Paralon is an option, and the only case where we can also show revenue.
Can Paralon Proof show the revenue the GPUs produce?
Only for capacity rented through Paralon, where we bill every minute ourselves. Revenue from the borrower's own customers is shown as declared by the borrower, next to what the machines show, so a contract for 64 GPUs with 20 at work stands out.
What does a certificate authority for GPUs mean?
A comparison with the web: a certificate authority vouches that a website is who it says it is, and the padlock in the browser shows it. Paralon Proof does the same for GPUs: it vouches, from the machines themselves, that the hardware behind a loan or a capacity claim is there and works, and the seal shows it. It is a description of what we do, not an official or regulatory status.
What is proof of silicon?
Our name for proof of reserves applied to hardware: evidence, from the machines themselves, that the GPUs a loan, a token or a capacity claim rests on exist and work. A balance sheet says the GPUs were bought; proof of silicon says they are still there today.
Can it be used onchain, for tokenized GPU loans?
Yes, that is part of the pilot: once a day, a facility's signed reports become one proof published on the chain the protocol's loans already use, with a feed per facility that a smart contract can read. Any single report can be checked against it.
Can the borrower fake it?
Faking one reading is possible. Keeping many readings consistent with each other, every minute, is far harder than writing a monthly report. Proof replaces trust between audits with checks; it does not replace the occasional physical audit.
What happens to the collateral if a borrower defaults?
The servers can be rented on the Paralon marketplace where they stand: by the minute, as spot capacity, or kept private for buyers the lender chooses. Idle collateral earns while the lender decides what to do with it.
Which GPUs are covered?
NVIDIA data center and workstation GPUs, B300, GB300, B200, H200, H100, A100 and RTX PRO 6000 among them.
Is Paralon Proof available now?
The measurement every Paralon node goes through is live today. The collateral view, matching to the financed list, signed reports and the onchain proof are being built with pilot partners. Write to [email protected] to join.