Peter Diamandis put the question plainly: if software agents are going to buy and sell things for us, why should they use USDC? Why not "Satoshi's" coin, bitcoin, or one of the tokens already created for agents?
The answer came from Nikhil Chandhok, Chief Product & Technology Officer at Circle, the company that issues USDC. He spoke on stage at Moonshots Live 2026 alongside Diamandis, founder of XPRIZE and Singularity University, Emad Mostaque of Intelligent Internet and ARK Invest's Cathie Wood. The session was recorded on September 25, 2026, and published as an episode of Moonshots with Peter Diamandis on September 29.
Chandhok made a case for a dollar-backed token and for a blockchain Circle built for payments. He also said where that system still falls short.
The basics
An AI agent is software that carries out tasks on a person's behalf, such as booking, buying or hiring. For agents to do business with each other, they need a way to pay.
A stablecoin is a digital token designed to hold a steady value. Circle says each USDC can be redeemed one-for-one for US dollars. The company says the coin is backed by cash and cash-equivalent reserves, mostly held in a government money-market fund managed by BlackRock, and that it publishes reserve information (Circle).
USDC lives on blockchains. A blockchain is a shared ledger that many computers keep in sync. On a public blockchain, anyone can see every transaction.
Four reasons for USDC
Track record. Chandhok said USDC has already settled more than $100 trillion across more than 30 public blockchains. Because every transaction on those chains is public, he called it "hardened infrastructure." He said Circle has created and redeemed close to a trillion dollars of USDC, though he first said "billion" before saying "trillion."
Always on, and cheap. Financial infrastructure has to run around the clock every day of the year, he said, and it has to be inexpensive: "You want to pay cents, not BIPs." BIPs are basis points, fees charged as a share of each transaction, as card networks do.
Regulation. Chandhok called the GENIUS Act "a great tailwind in the new year." The US law created a federal framework for stablecoins. He said it would take effect in January after an 18-month implementation period. Once it does, he said, businesses could hold stablecoins as cash or cash equivalents and use them for payments. The enacted law was signed on July 18, 2025. It takes effect on whichever comes first: January 18, 2027, or 120 days after federal regulators issue final rules (Public Law 119-27). January 2027 is therefore the latest start date, not the only possible one.
Instant settlement. When you swipe a credit card, Chandhok said, the merchant waits days for the money. During that time the merchant carries the risk that the payment never arrives. Stablecoins settle instantly, he said, so that risk goes away. Diamandis compared the shift to the telegraph disrupting the Pony Express.
How many transactions a second?
The conversation then turned to scale. A question from the stage cited Matthew Prince, the head of Cloudflare, as someone "very focused on agentic commerce, agentic payments," who believes the agent economy will need blockchains that handle 20 to 100 of something per second.
The units caused confusion. The first version was 20 to 100 transactions a second, which would already be within reach. The exchange then settled on 20 to 100 million, set against Nasdaq, said to peak at about 2 million per second. The question put Visa at roughly 20,000 per second, framed as a guess, and said Ethereum does not come close.
Cloudflare has announced plans for a dollar-backed stablecoin, NET Dollar, for agents that book flights or buy groceries (Cloudflare). It has also described a system that lets AI crawlers pay websites for access (Cloudflare blog). Neither announcement gives the throughput figure the question attributed to Prince.
Chandhok said Circle's chain, Arc, is already "orders of magnitude" faster than Ethereum, at tens of thousands of transactions per second. Circle's own launch announcement gives no transactions-per-second figure. He said reaching 2 million, 5 million or 10 million per second is still an open problem: "Those are hard technical questions still."
He expects solutions to follow demand, as they did when Ethereum scaled through extra systems built on top of it, called "layer twos." He said he doesn't yet know what design would work at that scale. His "dream" is to reach 100,000 transactions a second "without batting an eyelid" and then a million.
He tied the need to how much real buying and selling agents do. If billions of agents act for people, they must actually transact. "Agents acting without actually transacting is not really economic activity," he said. "That's just noise, right? That's spam." He added that he should call Prince to find out more.
The conversation also raised the point that around 20 million transactions a second, the speed of light becomes a constraint, because signals take time to cross the globe. The discussion suggested working around it with separate regions or insurance on settlements, on the view that "the economy finds a way." Chandhok ended with confidence: "We will deliver TPS. There's no doubt in my mind."
Why Circle built its own chain
Diamandis told Chandhok he was "building a financial infrastructure for an economy that doesn't yet exist." He asked what problems remained and why Circle had built its own blockchain. Chandhok said the two answers were connected, and he listed features he said banks and large companies need.
Payment finality. When two banks settle a payment, it has to be final. Chandhok said no other blockchain on the market offers that guarantee today. Circle's announcement says Arc was designed for deterministic settlement in under a second. That means a confirmed payment is meant to be complete, with no later window to challenge it.
Known validators. Validators are the computers that check each new batch of transactions and agree it is valid. Many public blockchains run on unknown validators. Chandhok said many institutions see that as a security risk; they worry, for example, about "North Korea being in the money flow." Arc uses a known group of validators. Circle's announcement lists BlackRock, DTCC, Mastercard, Visa and ICE among them.
Privacy. "It's not okay for the world to know how much money we have in the bank account," he said, or what a transaction is worth. Many privacy techniques are expensive to compute, he said, and a solution has to fit existing regulation, because "you can't blow up the existing regulatory frameworks just because you have a good idea." He described privacy as solved. Circle's announcement is more cautious: it says confidential balances and transactions were still being developed for release across the network.
Cost. Chandhok cited data from the previous day: a median transaction cost about 0.005 cents to settle on Arc and 89 cents on Ethereum. That makes Ethereum about 17,800 times more expensive. He said cost matters because efficient markets depend on constant trading: agents move value back and forth, find inefficiencies and remove them. Each of those moves has to be cheap. He said a public blockchain competes with the private databases companies already run, yet has to work at global scale and be auditable by anyone.
He said what still needs solving for agents is partly about the agents themselves. They need personhood, accountability for their mistakes, and a work history people can check before hiring them. He pointed to coding agents: unless you pay a monthly subscription such as $200, he said, you pay first and then wait for the result to come back.
Do banks disappear?
Diamandis called banks the "woolly mammoth in the room." He said the capabilities described cover much of what banks do today. "Do they know they're cooked?" he asked.
"I don't think they disappear," Chandhok said. He said his hope is that banks will work with Circle through the change. "Some people will transition with us and some won't. So I don't know if they're cooked."