TokenOS paid its customers with their own money

TokenOS AI (TOS) sold "GPU nodes" to the public and promised 27% to 40% every 30 days. It also launched the TOS token. The public ledger shows the returns came from new purchases, not from any computing business. Payouts stopped on August 12, 2026 and never came back. This is the record of the TokenOS AI scam.

What this site says

TokenOS AI (TOS) told customers their money bought GPU servers. The servers, it said, were rented out to paying compute customers through a system it called A²E, the "Arbitrage & Orchestration Engine." The returns came from that revenue, paid every 30 days. The ledger shows where the returns came from: other customers' purchases. Most of those purchases were payouts going straight back in. Nine of every ten dollars paid out were spent on a new purchase, often the moment they landed. No payment from a compute customer has been found reaching the payout accounts, and no payment to a data center has been found leaving them. On August 12, 2026, new purchases fell below what was due that day. The payouts stopped, and they have not restarted.

The operator came closest to agreeing once, in public, when a member called the arrangement "ponzinomics."

What about the bank?

TokenOS's answer to all of this is that the real business ran through a bank account, and the ledger cannot see it. Compute customers paid in dollars, the argument goes. The ledger shows only the crypto half. It is the most common objection to this site, so it deserves a straight answer.

The ledger cannot see inside a bank. It can see every dollar that crosses between the bank side and the crypto side. Money has to be converted to get onto the ledger, and every conversion arrives from an exchange or an on-ramp, a company that sells crypto for dollars, with a date and an amount. That is enough to test the claim.

• The first explanation, on August 13, 2026, was that this had happened before. Bank of America, TokenOS's official Telegram account said (this page calls it the account from here on), "tends to freeze our bank account" at "the end of every monthly epoch", and this time "they did it with our fiat-to-crypto onramp wires pending". About two hours later: "it happens every single month". The ledger shows no on-ramp wire reaching the treasury at the end of any month, or at any other time, apart from about $312,000 that came through a trading desk, most of it matching the operator's own ETH sales. Every big payout day before August was paid from purchases, or from customer money the operator had sent off Solana and brought back.
• Payouts were never funded from the bank side. For the whole life of the operation, every payout left the treasury in money that had come in as a purchase, or had gone out of the treasury and come back. If compute income existed, it was not what paid customers.
• The payout calendar was fixed in advance. A node paid out 31 days after purchase, 35 for a GB300, and TokenOS's own records showed exactly how much came due on each day, a month ahead. August 12 was one of the two largest payout days in the project's history, about $2.7 million across 460 nodes. A business with revenue sitting in a bank, knowing that a month in advance, converts the money early. On the previous big day, July 12, something like that did happen: about $514,000 reached the treasury from the operator's own Ethereum accounts on July 11 and 14. For August 12, nothing arrived, and the next day's explanation was that the wires were still pending.
• Converted bank money did reach the treasury, about $312,000 of it in June and July 2026. Four of those deposits, about $253,000, each arrived within a day of the operator's own account selling a similar amount of ETH for cash. That ETH was not revenue. It was customer money: it had left the treasury weeks earlier, gone off Solana through the bridges, and been turned into ETH on the way. So the only bank money that ever came in was the customers' own, going out one door and back in another. The dates and amounts are on How the money moved. Nothing else from outside arrived.
• When purchases fell short on August 12, the bank side did not step in. Five weeks later, asked how much had been converted for payouts, the account answered: "In this fiasco, none yet." Four hours later it said it would "continue on ramping $1-2M each day until this is resolved". No on-ramp deposit has reached the treasury since the stop. Ten weeks on, no regular payout has gone out.
• The operator described the arrangement itself. On August 24, 2026, a member asked why payouts now depended on a bank when the treasury had always paid them directly. The account wrote: "It always has. Compute buyers pay in fiat." Asked where the conversion from dollars showed up on the ledger, it wrote that "98% of the time it's just a matter of accounting categories", with "close to equal fiat on one side and USDC on the other."

• The claim is still being made. On October 19, 2026, a member asked whether all the money was still sitting in the company's bank accounts, all accounted for. The reply, nine seconds later: "Correct!" By that day the treasury on the ledger held about $0. It had sold its last SOL two days earlier. The most recent money into the weekly payment account, the day before, had come from a customer's wallet, not from any bank.

So the question is not whether a bank account exists. It is why a business with real revenue paid every customer out of other customers' money, and stopped the day that money ran short. A business with real revenue does not. A fraud does.

The bank side is also the easiest part of this to check. A bank statement is one document, and a regulator can obtain it with a subpoena. The operator could post it tomorrow. Everything on this site can be checked by anyone with a browser. The operator's account of the business can be checked by one piece of paper, and so far nobody outside the company has seen it.

Where the money went

Customers paid one account. This site calls it the treasury. Payouts left from the same account. The rest moved as the diagram shows. Boxes with a red outline are operator accounts; each was created or funded by another operator account. The full account list is on Verify it yourself, and the step-by-step record is on How the money moved.

Purchases against what came due

Purchases roughly doubled every month from March to June, and grew by half again in July. Much of that buying was the previous month's payouts coming back. The bill followed one month behind and grew faster, because every round added 27% to 40% to what was owed. In August about $17 million came due against $6 million of purchases. That is the mechanism in one picture. The figures come from TokenOS's own purchase records, valued as described on How the money moved.

What is owed

Payouts began to fail on August 12, 2026. The last regular payouts went out the next day. As of October 22, 2026, no regular payout has gone out since. Every node ever sold has now passed its payout date. The unpaid total is about $20 million.

That figure comes from TokenOS's own records of node purchases. It counts every node that passed its payout date on or after August 12 with no payout recorded. Each node is valued at what its GPU type paid before the stop.

TokenOS AI is run by two men. Tyler R. Gonty uses the name Tyler Ray on social media. He has also gone by Ty Gonty. Christopher Kuntz lists himself as its technology and security chief. Both are covered on Who runs it.

Timeline and promised dates

The record runs from Gonty's earlier company in 2006 to the payout stop on August 12, 2026. Since the stop, the TokenOS account has named a day for payouts 12 times. Every one passed with no payout. The newest is Monday, November 2. Both lists are on the Timeline page.