Since 2026, a concept deeply related to AI has appeared frequently in encryption research and investment institutions, namely Web4.0. Different from previously popular narratives, the focus of this discussion is not on new consensus mechanisms or higher throughput, but on the switching of execution subjects—from humans operating the Internet to autonomously operating AI Agents directly participating in economic activities.

GoToMars Research Institute has previously continued to track the direction of combining AI and encryption. This report starts from the actual progress of infrastructure construction and evaluates the real development stage, structural opportunities and realistic constraints of this track in April 2026.

1. Structural mismatch that has been ignored for a long time

Before discussing the Agent economy, we need to first understand a practical problem: the current AI model already has a considerable degree of perception and decision-making capabilities, but the right to act at the commercial level is almost completely missing. Every aspect of the traditional financial system, from account opening and identity verification to payment authorization and contract signing, is based on the real-name system of natural or legal persons. An AI program running in the cloud can neither legally open a commercial bank account nor independently sign a legally binding agreement. The result is that AI can only assist humans as a tool and cannot assume the role of an independent economic subject.

A deeper mismatch occurs at the level of payment granularity. Taking Stripe's standard rate as an example, a single payment requires a fixed cost of 2.9% plus $0.30. The scenario assumed by this pricing is a human consumption behavior with a large single amount and low frequency. However, the actual behavior pattern of Agent is completely opposite. A single API call may only cost 0.001 to 0.01 US dollars, but the call frequency is as high as dozens of times per second. Under such transaction characteristics, the fixed handling fee of traditional card organizations exceeds the transaction amount and cannot support commercial micropayment scenarios.下表对两类用户在几个关键维度上的差异做了梳理。

Table 1 Characteristic differences between human users and AI Agents in key interaction dimensions

This mismatch means that no matter how powerful the model is, AI can still only play the role of information processor under the existing financial system. The prerequisite for Agent to truly become an economic subject is to find a set of underlying protocols that jump out of the traditional financial identity framework.

2. Why encryption infrastructure has become an alternative path

The permissionless blockchain network provides a structural solution to the above problems. In networks such as Ethereum, Solana or TRON, generating an on-chain address only requires local code to complete the generation of a public and private key pair. The entire process is completed in milliseconds without any centralized approval. This means that the AI ​​Agent can create an independent identity carrier for itself or its derived subtasks at any time, and directly participate in on-chain transactions as an asset undertaker. The introduction of stable coins further solves the problem of value units. The current total circulation scale of USDT and USDC has exceeded 250 billion US dollars, providing a stable pricing and settlement medium. Coupled with smart contracts and composable DeFi protocols, complex financial operations such as fund scheduling, lending, and hedging can be completed through programmed interfaces and are no longer restricted by business hours and manual approval.

The attractiveness of this alternative path is the core reason why institutional investors have recently publicly been bullish on Crypto in the Agent economy. The core logic is that the certainty and verifiability of the encryption system are more in line with the needs of machine behavior - AI will not fatigue or forget, and can complete contract audits and address verification within seconds, which is exactly the weak link of human users for a long time. Under this narrative came Web 4.0, namely:

Web 4.0 = Web 3.0 + AI Agent

The underlying proposition of Web4.0 is thus established, which is to combine encryption infrastructure with autonomously running AI Agents to allow machines to move from the information layer to the economic layer.

3. Progress and constraints of the three key agreements

In order for Agent to truly become an economic entity, it must fill three protocol gaps on the original blockchain infrastructure: payment, identity and tool invocation. In the past twelve months, these three gaps have been filled by three representative agreements, and progress has been concentrated between the second half of 2025 and the first quarter of 2026.

Figure 2 Timeline of key nodes in Web4.0 infrastructure construction (May 2025 to April 2026)

The payment layer is filled by the x402 protocol released by Coinbase in May 2025, which reuses the long-term idle 402 Payment Required status code in the HTTP protocol, allowing the server to directly return payment requirements in an HTTP request, and the client can obtain resources after completing the payment through a stablecoin signature. On April 2, 2026, x402 Foundation officially joined the Linux Foundation to carry out open source governance. The Foundation is co-sponsored by Coinbase, Cloudflare, and Stripe, and more than 20 institutions including AWS, American Express, Google, Mastercard, Microsoft, Polygon Labs, Shopify, Solana Foundation, and Visa participate in governance.

The identity layer is filled by ERC-8004, which is dominated by the Ethereum ecosystem. This standard was jointly drafted by representatives of MetaMask, the Ethereum Foundation, Google and Coinbase, and the reference registry deployment was completed on the Ethereum mainnet on January 29, 2026.

The tool call layer is dominated by the MCP protocol, which was open sourced by Anthropic in November 2024, and was donated to the newly established Agentic AI Foundation under the Linux Foundation on December 9, 2025.

It is worth noting that the advancement of the protocol layer is significantly faster than the accumulation of actual commercial traffic. According to a report by Unchained on April 3, 2026, x402 has an average daily number of transactions on the Base network of approximately 55,000, and has processed a total of approximately 97 million transactions on the Base chain to date. Data from on-chain analysis agency Artemis shows that the average daily cross-chain transaction volume of x402 is approximately US$28,000, and the average single transaction is approximately US$0.20. At the same time, approximately half of the current transactions are estimated to be wash volume or testing behavior, and the real commercial traffic is still in the very early stages. Therefore, the time difference between protocol standards and commercial adoption is the most critical observation point at this stage.

Table 2 Progress and realistic constraints of Web4.0 three-layer key protocols

Source: Linux Foundation, Ethereum Foundation, Anthropic, The Block, Unchained public disclosure; compiled by GoToMars Research Institute. Data as of mid-April 2026.

4. Typical case: computing power layer - the first basic asset to be positively claimed by the capital market

While payment protocols and application scenarios are still in the early stages of commercial verification, the entire Web4.0 stack is the lowest-level computing asset, taking the lead in completing the connection with traditional financial channels. This is a phenomenon that is easily overlooked but has great signaling significance - traditional capital did not wait for the Agent application to run through before entering the market, but chose to establish asset exposure from the computing power layer first.

The most typical logo comes from Grayscale. It filed a Form S-1 registration statement for the conversion of Grayscale Bittensor Trust into a spot ETP with the SEC on December 30, 2025, intending to be listed on NYSE Arca under the stock code GTAO. On the same day, Bitwise also submitted registration applications (Form N-1A) for eleven strategic crypto ETFs to the SEC, including a product specifically tracking TAO - the Bitwise TAO Strategy ETF. On April 2, 2026, Grayscale further submitted a revised version of Amendment No. 1 to the above-mentioned S-1 to the SEC, pushing the product closer to official launch. In Switzerland, the Safello Bittensor Staked TAO ETP (code: STAO) jointly issued by Deutsche Digital Assets and Safello has been launched on the SIX Swiss Exchange on November 19, 2025. It is traded in US dollars, with 100% physical support and superimposed staking returns. This means that the European market has taken the lead in introducing TAO-related products to mainstream exchanges before the North American ETF has been finally approved.

Secondary asset allocation on the institutional side is also accelerating. Grayscale rebalanced its Grayscale Decentralized AI Fund on April 7, 2026, increasing the weight of TAO from 31.35% to 43.06%, becoming the largest change in the weight of a single asset in this adjustment, and no other assets were added or removed during this rebalancing. This change in weight indicates that institutions have begun to truly incorporate decentralized AI into the main channel of AI narrative. In the past, institutional expression of this theme was almost entirely completed through NVIDIA and a few large model manufacturers. TAO is the first crypto-native asset to be significantly reallocated by mainstream decentralized AI theme funds.

Scarcity on the supply side further reinforces the long-term logic of this trend. Bittensor completed its first halving on December 14, 2025, and the daily new issuance dropped from 7,200 to 3,600, maintaining a homologous design with the supply rhythm of Bitcoin (Grayscale Research, December 2025; Bittensor official documentation). With about 70% of the circulating supply being pledged for a long time and the limited surplus being absorbed by the treasury and ETP structures of listed companies, the circulation pressure on assets at the computing power layer is structurally tightened. This is the fundamental reason why institutions are willing to establish positions before the application layer.

But there is also a real risk that must be pointed out here. On April 10, 2026, Covenant AI, the leading subnet operator of the Bittensor network, announced that it would withdraw from the network and sell off approximately 37,000 TAO, which once triggered a price correction of more than 20% and raised questions about the centralization of network governance again. This incident reminds institutional funds that although the logic of capitalizing the computing power layer is established, it is still in the dual running-in period of governance structure and network stability, and cannot be simply compared to the early institutional acceptance path of Bitcoin.

The signal released by the computing power layer has two meanings. First, the opening of institutional channels means that node operators, pledge service providers, subnet developers, and compliance custodians built around this layer will receive structural dividends when funds enter, which is a direction that can be deeply involved in the next stage; second, when institutions bind TAO and decentralized AI as configurable assets, the valuation anchoring of the remaining layers of the entire Web4.0 stack will gradually be established, and the assetization path of the three layers of payment, identity, and application will also gain reference coordinates.

5. Several judgments of GoToMars Research Institute

Based on the above observations, the Web 4.0 track is currently in a stage of centralized infrastructure construction and relatively lagging commercial adoption. The three forces of narrative, capital and protocol have completed the initial gathering, but the real commercial traffic on the chain is still in the early stage of verification. For investment institutions and project parties, this stage has both opportunities and risks. The key lies in whether they can make accurate judgments on the misalignment of the agreement window and the commercial window.

Table 4 GoToMars Research Institute’s three judgments on the Web4.0 track

Source: GoToMars Institute, April 2026.

The first judgment is that the outcome of the standard game will be known within the next twelve months. Although x402 and ERC-8004 have achieved phased breakthroughs, the formation of real traffic will still take time. Once the standards are established, tool chains, SDKs and audit services built around them will usher in concentrated opportunities. The second judgment is related to the migration of product forms. The assembleable and programmable callable product architecture represented by Skills will gradually be copied from AI Trading to traditional industries such as actuarial actuarial, supply chain performance, and content copyright settlement. This direction is a vertical incubation opportunity worthy of focus in the next stage. The third judgment is that competition in the stablecoin clearing layer will become a key place for the actual implementation of the payment layer. TRON, Solana, Base, and BNB Chain are all competing for the minds of this layer. Projects to build comprehensive Agent financial operating systems around different stablecoin networks have strong structural opportunities.

The slot window for underlying infrastructure has begun to narrow, and the vertical application layer that can transform protocol capabilities into real commercial closed loops is an opportunity worthy of more attention between 2026 and 2027. The completion of the protocol layer determines the possible boundaries of upper-layer applications, but the real value ultimately depends on the service capabilities for machine users in specific scenarios.