Circle Launches ARC Blockchain: Transactions at a Millionth of a Cent, Built for the AI Agent Economy

Source: No Priors | Published: 2026-04-09T10:01:08Z

When AI agents need cross-border settlements in milliseconds and 5-cent microtransactions, Circle is building the financial infrastructure for the machine economy with fully-reserved stablecoins and its custom ARC blockchain.

The intellectual roots of stablecoins trace back to a 1930s economics debate

Allaire's starting point wasn't the Bitcoin whitepaper — it was the Austrian school of economics and an old proposal called the "Chicago Plan." After the Great Depression, economists led by Irving Fisher proposed a "full-reserve" system: government-issued currency must be 100% asset-backed, and banks cannot use it for fractional-reserve lending. The banking industry lobbied hard against it, and the eventual compromise was the FDIC — the risk stayed, they just added a layer of insurance.

For decades afterward, the leverage problem kept resurfacing. The 2008 financial crisis was fundamentally about 12x, 14x, 30x leverage stacked on top of base assets. Allaire's thesis: the dollar's status as the primary reserve currency will persist for another three, four, five decades, but we can build a safer version within that framework. Stablecoins are the digital form of full-reserve money, and last year's GENIUS Act enshrined this principle in federal law — you cannot leverage stablecoin reserves in any way.

USDC is backed by a Treasury portfolio with 13-day duration

Every USDC is backed by short-term U.S. Treasuries, overnight Treasury reverse repos, and a small amount of cash. The portfolio's average duration is roughly 13 days — extremely liquid, essentially functioning as a cash instrument. Custodians include institutions like BNY Mellon, which manages tens of trillions in assets, and BlackRock provides daily transparency disclosures.

The range of use cases is vast: on one end, a 25-cent digital item purchase in a blockchain game, or a 50-cent inference output trade between AI agents; on the other, the world's largest electronic trading firms settling multimillion-dollar capital markets transactions. Stripe and Shopify merchants use it. Visa uses it to replace legacy internal bank settlement rails. Ramp just integrated it into enterprise treasury systems. One crypto company raising funds required investors to send USDC instead of wire transfers for a simple reason — wires have afternoon cutoff times, and USDC settles on weekends too.

The agentic economy needs an entirely new financial infrastructure

Allaire believes we're witnessing "the most dramatic leap in fundamental technological capability in the past three months." His core thesis: a growing share of real economic output — especially in the knowledge economy and service delivery — will be performed by AI agents. Agents will collaborate, consume each other's services, and purchase specialized intelligence outputs.

Why can't the existing financial system handle this? He listed several hard constraints: it can't interoperate globally, can't settle instantly, can't be called programmatically by arbitrary software, can't let agents dynamically create financial endpoints, can't scale to billions or trillions of transactions, and can't process 5-cent or 10-cent microtransactions. None of this was feasible before third-generation blockchains. Now USDC's transaction volume is growing far faster than its monetary base — a sign of accelerating money velocity — while transaction costs have dropped to sub-cent levels.

ARC: an "economic operating system" built for the machine economy

Circle is launching its own blockchain, ARC, positioned not as another general-purpose chain but as an "economic operating system." Three key differences from chains like Solana and Ethereum:

Known validator set. ARC's infrastructure operators are large financial infrastructure companies (the specific roster hasn't been announced), not anonymous nodes. This enables guarantees that "bad actors aren't processing your transactions," plus deterministic settlement finality — transactions can't be hard-forked or reorganized, with final settlement in hundreds of milliseconds.

USDC as the native token. No volatile gas token. Transaction fees are denominated in real dollars. For enterprises, it's as intuitive as paying an AWS bill — budgeting, compliance, and financial workflows plug in directly. Transaction costs on ARC can run as low as one-millionth of a cent.

Built-in privacy primitives. Enterprises don't want their on-chain activity visible to everyone, and individuals don't want to get doxxed. ARC integrates zero-knowledge proof-based privacy from day one while preserving compliance audit capabilities.

Allaire's vision for the agentic economy goes beyond payments. He believes AI agents need a trusted intermediary layer to instantiate entities, store value, and execute and schedule contracts — all requiring real-time, mathematically provable computational integrity. Blockchain's tamper resistance, full auditability (all inputs and outputs are publicly verifiable), and computational integrity guarantees are exactly what this scenario demands. He even argues that future "companies" may themselves be on-chain software machines — with novel governance and contract structures, blending human and AI agent participants.

Emergent cooperation between agents is already showing early signs

Allaire mentioned that projects like OpenClaw and MoltBook (phonetic) have demonstrated a fascinating phenomenon: emergent cooperation, interaction, and coordination between AI agents. While still bleeding-edge, the direction is clear — AI agents from around the world, built on different models and LLMs, need a trusted coordination space. In that space, they can create entities, store value, and schedule contracts, with all operations verifiable in real time.

This is different from the speculative narratives of earlier years. Allaire draws an analogy to the internet's "broadband moment" — a decade-plus of grinding through the desert, with Wi-Fi, broadband, and new devices falling into place one by one, until it finally became possible to deliver software, media, and real-time communication over the internet. Blockchain has traveled roughly the same road, and now societal demand, financial system needs, and the AI agentic economy are all converging at the same point in time.

Asset tokenization is happening at every layer simultaneously

"When will I be able to buy fractional shares on-chain?" — it's already happening. Allaire shared a data point: the most actively traded tokenized stock on-chain isn't Tesla, isn't an S&P index fund — it's Circle itself.

The deeper shift is that every layer of the financial system is pushing tokenization forward at once: the base layer of equity registrars (like Computershare), the middle layer of depository and clearing systems (like DTCC), and the upper layer of exchanges and brokerages (NASDAQ, NYSE). The SEC issued explicit operational guidance about a month ago, defining obligations at each layer.

Current growth in tokenized equities is primarily driven by users outside the U.S. who previously lacked convenient access to American stocks. But Allaire thinks the more interesting question isn't "putting TV shows on the internet" — it's "what can you do on the internet that TV never could?" — fractionalization, novel lending structures, portfolio bundling, with AI potentially playing a major role. Prediction markets are evolving in this direction too. Polymarket runs on USDC, and its most active users are, tellingly, market makers from traditional finance — they mine prediction markets for "what is reality" signals, then map those back to equities and derivatives.

Replacing mining energy consumption with inference compute: a direction worth watching

Several recent papers have explored an idea: binding Bitcoin's proof-of-work mechanism to GPU inference computation. Bitcoin's proof-of-work fundamentally consumes energy — the byproduct of all that compute is heat, nothing else. If inference computation itself serves as the "work," then compute simultaneously produces useful intelligence output and can serve as the consensus basis for a cryptocurrency.

Allaire finds this compelling and said something you don't hear often: "I don't know what we'll be using ten years from now. Bitcoin has been around a long time, but the paradigm shift in energy infrastructure, the efficiency of energy-to-intelligence conversion, new computational layers — these open a channel that didn't exist fifteen years ago." That may sting for Bitcoin maximalists, but coming from someone who started from the Austrian school of economics, he's clearly reassessing the definition of "sound money."

Double-digit GDP growth in the 2030s is "not unrealistic"

Asked about AI's impact on the global economy, Allaire admitted he's not entirely sure what to think. But he offered a range: double-digit GDP growth rates in the 2030s are "not unrealistic" for much of the world.

He immediately added an important caveat, though: GDP growth itself may no longer mean what it used to. The biggest risk is GDP growth becoming a game of "capital capturing more capital" while humans get cut out. This loops back to a theme he returns to repeatedly — the need for a new social contract.

He draws an analogy to the Enlightenment and the Industrial Revolution: every major technological transformation has been accompanied by a redefinition of the social contract, which in turn reshaped social, political, and economic organization. The AI-driven transformation will be no different, and there will be an uncomfortable lag between disruption and the emergence of new institutions. His bet is that on-chain organizations — entities that blend human and AI agent participants with novel governance and contract structures — may become the most productive corporate form in the history of human economics.