The labs said slow down.
Compute
Belongs to
the people.
NCOMP is a token whose trading fees buy GPUs. The GPUs run open models. Holders use them free and share the revenue.
Don’t nationalize the labs. Nationalize compute.
Local AI is not optional anymore. Own the machines.
Why tonight
The labs are pausing.
Frontier labs have publicly said progress should slow down. A pledge is not a limit.
The models they build keep getting shipped, and the hardware they run on keeps getting bought.
Open weights are free. GPUs aren’t.
Anyone can download an open model and read its weights. Almost nobody can afford the machines that serve it.
Inference is priced by whoever owns the rack, not by whoever owns the model.
So the public should own the GPUs.
Compute is the bottleneck, and it is being built out in private hands. Local capacity is how people keep a say in what runs.
NCOMP buys the machines in the open and lets the holders use them.
3%
every trade buys hardware
40%
of inference revenue to holders
100%
public supply, no team allocation
0
buybacks
The reserve, live
Reserve cash · NVDA
—
Pre-launch
Reserve cash · USDG
—
Pre-launch
Hardware · book
—
Pre-launch
GPUs online
—
Pre-launch
Distributed · USDG
—
Pre-launch
How it works
- 01 A trade pays 3%.
- 02 90% of the fee goes to the Reserve, 10% to Pons — 2.7% and 0.3% of the trade.
- 03 Fees accrue in NVDA.
- 04 The Reserve spends them on GPUs: bought in the open market, owned outright, racked in a colo.
- 05 The GPUs sell inference in USDG on open models, and serve holders free.
- 06 Inference revenue splits 60 / 40: 60% reinvested in hardware, 40% paid to holders in USDG. Trading fees are never distributed.
For holders
Hold 100,000 NCOMP for the basic tier; 1,000,000 for the full tier. Balances are read on-chain hourly.
40% of inference revenue is paid to holders in USDG, pro-rata by balance snapshot, claimable from the Distributor.
No staking, no lockup. Holding is enough.
The hardware
| Node | GPUs | Location | Acquired |
|---|
Hardware is held by a protocol LLC and racked in a colo. Serial numbers and invoices are published; book value is written down straight-line over 36 months.
Tokenomics
1,000,000,000 NCOMP. Fixed supply, no mint function.
100% sold on the public curve. No team allocation, no pre-mint.
3% on every buy and sell: 90% of the fee to the Reserve, 10% to Pons.
That is 2.7% of the trade to the Reserve and 0.3% to the venue.
Fees are spent on GPUs only — never bought back, never distributed.
Addresses
| Contract | Address | |
|---|---|---|
| Treasury | Published at launch |
Blockscout |
| FeeVault | Published at launch |
Blockscout |
| Distributor | Published at launch |
Blockscout |
| NCOMP | Published at launch |
Blockscout |
| NVDA pair | 0xd0601CE157Db5bdC3162BbaC2a2C8aF5320D9EEC |
Blockscout |
Questions
What backs NCOMP?
Cash and hardware. The Treasury holds NVDA and USDG; the Reserve owns GPUs outright. NAV is cash plus hardware at depreciated book value, published monthly. There is no other backing.
How do holders earn?
40% of inference revenue is distributed to holders in USDG, pro-rata by balance snapshot, claimable through a Merkle Distributor. Trading fees are excluded: they accrue in NVDA and buy GPUs.
What are the free tier thresholds?
100,000 NCOMP for the basic tier and 1,000,000 NCOMP for the full tier, read on-chain hourly. No staking and no lockup.
Where do the fees go?
Every buy and every sell pays 3%. 90% of each fee goes to the Reserve and 10% to Pons — 2.7% and 0.3% of the trade. Fees buy hardware, and only hardware.
How do I verify?
Read the chain. Robinhood Chain has chain id 4663; the explorer is robinhoodchain.blockscout.com. The addresses above are published with Blockscout links.