Author: Tim Sun

In the past couple of years, discussions surrounding Ethereum have fallen into a strange divide: on one side, applications are sluggish, narratives have aged, and price performance seems rather flat; on the other side, core researchers continue to advance work in areas such as privacy, verification, and governance, which appear slow and are not widely understood by the public.

Only recently, with the emergence of ERC-8004 and Vitalik Buterin's systematic explanation of the relationship between AI and Ethereum, has this long-term context begun to take shape. If discussions prior to 2024 mostly revolved around using Crypto to hype an AI bubble; now, Ethereum has quietly entered the stage of building infrastructure for the AI era.

The question is not whether Ethereum can run large models, but a more fundamental macro proposition: In an AI era characterized by extreme concentration of computing power and exponentially expanding capabilities, what role should Ethereum play?

1. How Vitalik Views AI: What Is the Issue with Technological Accelerationism?

Vitalik's judgment and concerns about AI stem from the fact that the current development path of AI is being dominated by a form of indiscriminate technological accelerationism. The core logic here is not complex: the stronger the model, the better; the faster the progress, the safer; the leaders should win everything.

However, in Vitalik's view, this seemingly value-neutral inclination comes with a very clear power direction. To put it simply, under the framework of technological accelerationism, AI development will inevitably lead to: continuously enhanced capabilities, but control will persistently concentrate among a few giants.

Especially when one calls upon AI more frequently, it may not necessarily improve one's situation; instead, it may amplify the asymmetry of power. For instance, for individuals, AI makes profiling, surveillance, manipulation, and behavior prediction cheaper and more covert; for certain organizations, the threshold for large-scale content generation, information manipulation, fraud, and opinion guidance drastically decreases.

From this perspective, in Vitalik's view, AI may force humans into retirement and could progressively exclude them from decision-making and value distribution.

Thus, in his latest writing, Vitalik emphasizes explicitly at the beginning that the problem with AI is not computing power, nor is it model size, but the choice of direction. He does not oppose AI, but he opposes blindly accelerating without constraints or calibration mechanisms. This is also the reason he attempts to introduce Ethereum into the core discussion of AI:

Not to participate in the computing power race, but to provide a direction for technological evolution in the AI era that is not so easily consumed by power structures.

2. Why ERC-8004 Is Underrated: What Is the Value of Minimizing Verification of AI?

ERC-8004 was formally proposed in August 2025, jointly promoted by the Ethereum Foundation, the MetaMask AI team, Google, and Coinbase, with a goal of launching on the mainnet in the second quarter of 2026. In terms of importance, this proposal is one of the most crucial but easily underestimated in recent years for the Ethereum mainnet.

Because ERC-8004 does not solve the superficial problem of putting AI on-chain, but rather addresses a more fundamental and, paradoxically, counterintuitive problem: How can we use AI, which we do not trust, when it begins to independently execute economic actions?

Understanding ERC-8004 fundamentally involves understanding how Ethereum and Vitalik Buterin wish to incorporate AI into a verifiable, constrained, and exitable system. ERC-8004 has one important assumption: that AI is not worthy of being trusted or is assumed to be unreliable. Even under such circumstances, how can humans still collaborate with it under controllable risks? Design-wise, ERC-8004 is extremely restrained, only introducing three minimally necessary components: identity registration, reputation registration, and verification registration.

Put simply, this means giving AI a traceable identity, a visible behavior record, and a trusted evaluation that can be authoritatively verified in critical scenarios (such as DeFi).

Before ERC-8004, AI operated within platforms where the rules were set by the platform; ERC-8004 tries to turn this around; AI is just the executor, not the judge.

This also explains why it deserves to be discussed separately: ERC-8004 is not just a standard; it represents a new main thread Vitalik has found for Ethereum in the AI era. Ethereum is not seeking to participate in AI's capacity race but aims to become a trust-minimized foundation for AI economic actions. This is precisely what Ethereum has excelled at in the past, and it is why Vitalik repeatedly emphasizes the importance of decentralization and censorship resistance.

3. Don't Trust, Verify: How Can Privacy Technology Support AI?

Another direction that Vitalik repeatedly highlights is the new importance of privacy technology (especially ZK, zero-knowledge proofs) and verification logic in the AI era.

In Vitalik's eyes, this logic is very simple: if blockchain only provides transparency, then for AI, it is equivalent to running naked in a dark forest without any cover. If every RPC call from AI and every flow of funds across protocols is made public, then AI's strategies will be instantly preempted, and its intentions will be completely deconstructed. Without privacy, AI cannot become a true economic entity, nor can it operate sustainably on the blockchain.

Therefore, the privacy roadmap of Ethereum, including Vitalik's latest explanation, actually constructs a complete set of privacy solutions. For example, one can use TEE to encapsulate the entire RPC query, reducing the visibility of the query content to nodes or service parties; in the long term, it will gradually transition to PIR (private information retrieval), ensuring that cryptographically, nodes do not know what you queried;

For instance, zk-based payments can avoid identity relationship leakage in the payment path; one can prove payment has been made without exposing one's identity association and behavioral trajectory.

Additionally, there is the locally-based large model toolchain that keeps the most sensitive personal interactions and strategies local; of course, the most important aspect is client-side verification, enabling terminals to have verification capabilities. Even if computing power is accessed remotely, you can determine locally whether it has executed according to the rules.

Piecing these clues together, you'll find that in the AI era, the Ethereum Vitalik wishes to promote is not heading towards more transparency, but towards a verifiable yet inscrutable advanced stage.

In other words, it is about using cryptographic mathematical constraints to ensure the trustworthiness of AI's behavioral results and rule execution while allowing the AI's decision-making paths, attention data, and funding sources to remain necessarily confidential.

4. Towards a New AI Era: Where Is Ethereum Headed from Being a World Computer?

At this point, we can answer the initial macro proposition: What exactly is Ethereum laying out?

For the past decade, the core narrative of Ethereum has been that of a world computer. However, in the AI era, this metaphor may no longer be accurate and could be somewhat misleading, as blockchain will never outperform Nvidia's GPU clusters in terms of computational efficiency.

Taking into account ERC-8004 and Vitalik's latest thoughts, we may consider that Ethereum is completing a strategic upgrade from a computational layer to an AI economy and verification layer. In a context where AI continues to amplify capabilities and power, what is the most irreplaceable aspect is not efficiency; there will always be larger data centers and smarter models, but one cannot bypass or replicate the validation of AI.

What Ethereum is attempting to carry is precisely this layer. Verifying whether the actions of AI happen according to rules; providing a controllable, accountable economic structure for AI cooperation; allowing humans a certain degree of choice in the face of AI.

This path is destined to be slow and is bound not to be popular; yet, from the history of Ethereum over the past decade, it seems it has always reached today through such a commitment against speed.

In the entire wave of AI, Silicon Valley venture capitalists and tech giants mostly ask: just how intelligent and fast can we push AI to be?

What Ethereum and Vitalik are attempting to answer is another, more dangerous yet significant question:

When AI becomes inevitable, how much choice can humanity retain?

This, perhaps, is what Ethereum and Vitalik truly aim to lay out.