176 Million Transactions, $73 Million: Grayscale's AI-and-Blockchain Thesis and the Rebuttal That Landed Eight Days Later

The Three Demand Areas Grayscale Identified

Pandl positioned artificial intelligence and public blockchains as complementary, then identified three areas where AI adoption would generate blockchain demand. The first is agentic finance. To act on a user's behalf, AI agents need programmable wallets that can hold and deploy capital without intermediaries, and that activity will drive demand for micropayments, instant cross-border settlement, and automated trading and risk management. He named Ethereum and Solana as the networks suited to this.

The second is verifiable record keeping for computation, identity and reputation. As enterprises delegate more decision-making to AI, they need to verify which model, data and rules produced a judgement, while users need to distinguish humans from AI agents online. Here he cited Worldcoin's identity service as an example. The third is decentralised AI infrastructure, offering Bittensor as a counterweight to the concentration of compute and control among frontier labs and hyperscalers.

This Is Research Written by an Issuer

One premise should be established before reading further. Grayscale offers investment products tied to all four networks it named, and the report was published by the firm's investment research arm rather than an independent analyst. When Grayscale published a report on Zcash on August 20, overseas coverage explicitly noted that the firm operates the Grayscale Zcash Trust and therefore holds a direct financial interest in that asset's price and adoption, and that the note came from its research arm. The same structure applies here. Separate from the merits of the analysis, the selection of named assets should be read as potentially interested.

The Rebuttal Eight Days Later

On August 19, Max Wadington, Senior Research Analyst at Fidelity Digital Assets, published a report setting out six structural risks to this thesis. The first is that AI agents may not converge on public blockchains at all: closed systems operated by large technology and fintech firms could absorb the same activity, with advantages in performance, cost, user experience and regulatory clarity. The second is that payment volume may not translate into token value. Wadington argued that the primary economic beneficiaries of payment-driven growth may be stablecoin issuers and adjacent service providers rather than the underlying blockchain networks. The remaining four concern the limited economic value of increased AI-generated software output, the weakening of technical differentiation between networks as code becomes cheaper to produce, growing security threats, and regulatory constraints.

His central sentence runs: even if AI drives a substantial increase in overall digital economic activity, there is no guarantee that public blockchains will capture a meaningful share of it. The report also notes that most agentic payments today already execute on Layer 2s and higher-throughput networks, further compressing value capture at the base layer.

What the Numbers Show

The dispute is partly settled by data. According to research firm Keyrock, AI agents settled more than $73 million across roughly 176 million blockchain transactions in the year through April. That works out to approximately $0.41 per transaction. The count is enormous; the economic size of each transaction is minute. This does not demonstrate that the rails are unused — quite the opposite, it demonstrates they are being used exactly as intended, for micropayments. Which is precisely why Fidelity's point carries weight: at the same fee rate, transactions averaging under a dollar do not accumulate into meaningful revenue.

The Contest Has Already Moved Beyond the Settlement Rail

What neither report treats fully is the layered standards architecture that has come together over the past year. On July 14, 2026, the x402 Foundation formally launched under the Linux Foundation with 40 members including Visa, Mastercard, American Express, Stripe, AWS, Google, Cloudflare, Circle and Coinbase. x402 revives the HTTP 402 status code so that an AI agent can complete a payment inside an ordinary web request.

The crucial point is that these efforts are layers rather than competitors. x402 handles payment execution, Google's AP2 handles delegated authority and proof of purchase intent, and the Agentic Commerce Protocol from OpenAI and Stripe handles the checkout flow. The card networks, meanwhile, have deliberately positioned themselves as protocol-agnostic so that whichever standard prevails, their credentials sit underneath it. Visa's Trusted Agent Protocol and Mastercard's Agentic Tokens are both designed to occupy the credential and fraud-signalling layer beneath the rail rather than the rail itself.

In other words, the point of value accumulation in an AI agent economy is shifting away from where money moves and toward where it is proved that a payment reflects a legitimate delegation of the principal's intent. Of Grayscale's three areas, the second — verifiable records — is the most fundamental, but crypto projects are not the only contenders there. Incumbents holding existing credential infrastructure are further ahead on regulatory readiness and merchant distribution.

[Business Development Insights]

1. AI agent payments should be evaluated on fee density, not transaction count. A record of 176 million transactions in a year averaging $0.41 each shows that conventional payment-business metrics do not transfer. A revenue model premised on a rate applied to transaction value cannot recover fixed costs across hundreds of millions of sub-dollar transactions. Any institution planning in this space must decide at the outset either to build an architecture whose marginal cost per transaction approaches zero, or to charge at a layer above settlement rather than on settlement itself. Plans that leave this choice unresolved while committing to "enter AI agent payments" will fail on economics.
2. The revenue opportunity is in proving delegated authority, not in executing payments. It is telling that the card networks declined to compete on rails and instead took position in a protocol-agnostic credential layer, choosing a place where their credentials sit underneath whichever standard wins. The same structure applies to Japanese financial institutions. In a world where AI agents transact on customers' behalf, someone must prove and audit that an instruction genuinely reflects the principal's intent and stays within the scope of delegation — and that is the role of parties with an existing track record in identity verification and credit. Applying those assets at the attestation layer builds a more defensible position than constructing wallets or settlement infrastructure in-house.
3. Reading issuer-published research should be institutionalised as a process. The Grayscale-Fidelity disagreement is less a question of who is correct than a structural observation: both firms reached conclusions consistent with their own product line-ups. Grayscale offers products in the four named networks; Fidelity's franchise is centred on Bitcoin. Materials prepared for senior management should state explicitly whether the publishing party holds an interest in the asset discussed, and present it alongside corroboration from independent research houses or on-chain data. As the Keyrock measurement illustrates, adding a single figure that lets both arguments be tested materially improves the quality of the judgement.

[Sources]

• CoinPost, "Grayscale Analyses Rising Blockchain Demand From AI Adoption" https://coinpost.jp/?p=730791
• Grayscale Research (Zach Pandl), "Why AI Needs Blockchain Technology" https://www.grayscale.com/the-stack/why-ai-needs-blockchain-technology
• Fidelity Digital Assets (Max Wadington), "AI's Potential Impact on Digital Assets" https://www.fidelitydigitalassets.com/research-and-insights/ais-potential-impact-digital-assets
• CoinDesk, "AI could supercharge crypto but there's a catch, Fidelity Digital Assets says" https://www.coindesk.com/tech/2026/08/19/ai-could-supercharge-crypto-but-there-s-a-catch-fidelity-digital-assets-says
• BeInCrypto, "Fidelity Digital Assets Names 6 Risks to Crypto's AI Agent Thesis" https://beincrypto.com/fidelity-grayscale-ai-blockchain-agent-risk/
• crypto.news, "Fidelity names 6 risks to crypto's AI agent thesis" https://crypto.news/fidelity-names-6-risks-to-cryptos-ai-agent-thesis/
• The Crypto Times, "Grayscale Says AI Could Drive New Demand for Public Blockchains" https://www.cryptotimes.io/2026/08/11/grayscale-says-ai-could-drive-new-demand-for-public-blockchains/
• The Crypto Times, "Grayscale Flags Zcash Privacy Use as AI Expands Financial Data" https://www.cryptotimes.io/2026/08/20/grayscale-flags-zcash-privacy-use-as-ai-expands-financial-data/
• CoinDesk, "Visa, Stripe and Google join massive open-source project to let AI agents pay each other" https://www.coindesk.com/business/2026/07/16/ai-payments-have-a-new-open-standards-body-its-aim-is-to-reinvent-the-internet
• Mastercard, "Mastercard launches Agent Pay for Machines" https://www.mastercard.com/global/en/news-and-trends/press/2026/june/mastercard-launches-agent-pay-for-machines.html
• ForkLog, "Fidelity Assesses AI's Impact on the Crypto Market" https://forklog.com/en/fidelity-assesses-ais-impact-on-the-crypto-market/