---
title: "Vitalik Buterin Says DeepSeek V4 Could Power Ethereum’s Private AI Future Without Centralised Servers"
type: "News"
locale: "en"
url: "https://longbridge.com/en/news/288021711.md"
description: "Vitalik Buterin emphasizes the importance of DeepSeek V4 in enhancing Ethereum's privacy through local AI. He argues that current AI systems often rely on centralized infrastructure, risking user data exposure. Buterin's CROPS AI concept aims to create a secure, decentralized AI ecosystem. DeepSeek V4's capabilities allow users to manage blockchain tasks privately, but hardware limitations persist, with performance varying significantly across platforms. He advocates for AI models tailored for Ethereum to bolster security and privacy in blockchain interactions."
datetime: "2026-05-29T06:31:40.000Z"
locales:
  - [zh-CN](https://longbridge.com/zh-CN/news/288021711.md)
  - [en](https://longbridge.com/en/news/288021711.md)
  - [zh-HK](https://longbridge.com/zh-HK/news/288021711.md)
---

# Vitalik Buterin Says DeepSeek V4 Could Power Ethereum’s Private AI Future Without Centralised Servers

Ethereum’s push for stronger privacy protections is now becoming closely tied to the rise of local artificial intelligence, with Vitalik Buterin pointing to DeepSeek V4 as a key piece of infrastructure for the network’s future access layer.

The Ethereum co-founder says advanced AI systems are becoming capable enough to manage wallets, interact with blockchains, and potentially handle sensitive financial activity on behalf of users.

But he argues that today’s AI ecosystem is still heavily dependent on centralised infrastructure that quietly exposes user data.

Speaking about his CROPS AI vision, Buterin warned that many so-called local AI systems are not truly private despite running on personal devices.

CROPS AI stands for Censorship-Resistant, Open-Source, Private, and Secure AI.

> Updates since then:
> 
> \* Deepseek v4 is out. There \*is\* a 2-bit quant that can run within 90 GB ( https://t.co/X3AFAsiH02 ), and it works, however it's only fast on Apple hardware (I've head ~35 tok/s). On AMD, it's ~7 tok/s. IMO actually taking the effort to properly support more… https://t.co/zo04n5Cx0F
> 
> — vitalik.eth (@VitalikButerin) May 27, 2026

## Why Buterin Says Most Local AI Is Not Really Private

According to Buterin, many open-source AI tools still rely on external services from companies like OpenAI and Anthropic whenever they encounter tasks they cannot process independently.

He referenced the growing ecosystem around local models such as DeepSeek V4 and the Qwen 3.5 series, along with locally hosted agent frameworks.

However, he said many of these systems still “ping” corporate servers behind the scenes, exposing metadata, IP addresses, and potentially sensitive blockchain activity.

Buterin said the mainstream open-source AI ecosystem remains heavily focused on capability and convenience rather than privacy and security.

He first introduced the CROPS AI concept during the ETH Mumbai conference on 12 March 2026, where he described AI as one of the next major security risks facing crypto if the infrastructure remains dependent on centralised providers.

> At ETH Mumbai, Vitalik Buterin shifts focus from scaling to AI risks for crypto users. Introducing CROPS AI concept, he challenges Ethereum to pioneer AI privacy. Disciplined minds see beyond the hype to grasp the true revolutionary path. https://t.co/E9cEqVES70
> 
> — joshua.sol (@GuyGrill98439) March 13, 2026

## DeepSeek V4 Emerges As Ethereum Privacy Tool

Buterin highlighted the newly released DeepSeek V4 as proof that fully local AI is becoming practical on consumer-grade hardware today rather than years away.

He specifically pointed to a 2-bit quantized version of the model that can run within roughly 90GB of VRAM or unified memory.

> 👤 Vitalik says
> 
> > Vitalik Points to DeepSeek V4 as Local AI Meets Ethereum Privacy
> > 
> > Ethereum co-founder Vitalik Buterin updated his progress on local AI, saying DeepSeek V4 now has a 2-bit quantized version that can run within about 90 GB of VRAM, reaching roughly 35 tokens per second on Apple… pic.twitter.com/Ybyp3NRVhZ
> > 
> > — Wu Blockchain (@WuBlockchain) May 28, 2026

According to him, this creates a path for users to process blockchain-related tasks privately on their own machines without exposing wallet activity to remote providers.

Buterin said,

> “One other thing that has been on my mind is that there’s actually a lot of intersection between “CROPS Ethereum access layer” and “CROPS AI”. For example, we want a ZK way to make (paid) calls to remote LLMs. But if we have this, then it’s just as useful for solving another problem: private RPC reads in Ethereum.”

He explained that locally hosted AI systems could allow users to query Ethereum balances, smart contract states, and network activity without revealing their identities or transaction history to centralised RPC services.

The approach would rely on self-hosted infrastructure combined with zero-knowledge proofs, allowing blockchain interactions to remain private until users choose to broadcast transactions publicly.

Buterin described this as creating a “cryptographic sanctuary” where financial intentions stay entirely on a user’s own hardware.

## Hardware Limits Still Divide The Local AI Ecosystem

While Buterin sees major progress in local AI performance, he also pointed to large hardware differences across platforms.

According to him, the DeepSeek V4 setup currently achieves around 35 tokens per second on Apple hardware, while AMD GPUs only manage roughly seven tokens per second under similar conditions.

He argued that true decentralised AI must support multiple hardware vendors rather than optimising for a single ecosystem.

Buterin wrote,

> “IMO actually taking the effort to properly support more than one hardware manufacturer is a great example of the difference between mere ‘decentralized AI’ and genuine ‘CROPS AI’.”

He added that users would likely need at least 96GB to 128GB of unified memory or VRAM to comfortably handle the model’s 90GB quantization overhead.

Buterin also pointed developers toward optimisation work involving DeepSeek V4 Flash patches for AMD systems as an important next step.

## Ethereum-Tuned AI Models Could Help Secure Smart Contracts

Beyond privacy, Buterin believes locally hosted AI could become a security tool for Ethereum developers themselves.

He called for application-specific AI models trained specifically for Ethereum-related coding, auditing, and protocol verification tasks.

> ⚡ INSIGHT: Vitalik Buterin says local AI is advancing fast, highlighting DeepSeek V4’s ability to run on consumer hardware while pushing for Ethereum-integrated AI tools focused on privacy and smart contract security. pic.twitter.com/60jkPfNR9J
> 
> — Coinwaft (@coinwaft) May 28, 2026

Referencing models such as Mistral’s Leanstral, which reportedly runs within 70GB of VRAM while achieving around 38 tokens per second on AMD hardware, Buterin said specialised AI systems could help developers identify flaws in smart contracts before deployment.

He said,

> “We should have models fine-tuned for Ethereum-related use cases as well.”

He tied this directly to his broader push for formal verification and safer protocol development across the Ethereum ecosystem.

Buterin also highlighted projects such as Lucebox Hub and VoxTerm as examples of a wider shift toward user-controlled AI infrastructure that avoids dependence on third-party servers.

The broader goal, according to Buterin, is to revive a more cypherpunk-style internet model where AI systems act on behalf of users without leaking their identities, financial behaviour, or personal data to large corporations.

### Related Stocks

- [ETHE.US](https://longbridge.com/en/quote/ETHE.US.md)
- [ETH.US](https://longbridge.com/en/quote/ETH.US.md)
- [BLKC.US](https://longbridge.com/en/quote/BLKC.US.md)
- [ETHV.US](https://longbridge.com/en/quote/ETHV.US.md)
- [EETH.US](https://longbridge.com/en/quote/EETH.US.md)
- [IBLC.US](https://longbridge.com/en/quote/IBLC.US.md)
- [SATO.US](https://longbridge.com/en/quote/SATO.US.md)
- [BLOK.US](https://longbridge.com/en/quote/BLOK.US.md)
- [EZET.US](https://longbridge.com/en/quote/EZET.US.md)
- [ETHW.US](https://longbridge.com/en/quote/ETHW.US.md)
- [BKCH.US](https://longbridge.com/en/quote/BKCH.US.md)
- [CETH.US](https://longbridge.com/en/quote/CETH.US.md)
- [BTF.US](https://longbridge.com/en/quote/BTF.US.md)
- [ETHA.US](https://longbridge.com/en/quote/ETHA.US.md)
- [FETH.US](https://longbridge.com/en/quote/FETH.US.md)
- [QETH.US](https://longbridge.com/en/quote/QETH.US.md)
- [XEY.US](https://longbridge.com/en/quote/XEY.US.md)
- [ETHHKD.VAHK](https://longbridge.com/en/quote/ETHHKD.VAHK.md)
- [ETHUSD.VAHK](https://longbridge.com/en/quote/ETHUSD.VAHK.md)
- [AAPL.US](https://longbridge.com/en/quote/AAPL.US.md)
- [AMD.US](https://longbridge.com/en/quote/AMD.US.md)
- [OpenAI.NA](https://longbridge.com/en/quote/OpenAI.NA.md)

## Related News & Research

- [SBET: Revenue soared but net loss deepened on unrealized ETH losses; major ecosystem investments launched](https://longbridge.com/en/news/295401645.md)
- [Fidelity Ethereum Fund Expands Ether Staking With Anchorage Digital, BitGo, and Revised Sponsor Terms](https://longbridge.com/en/news/295438250.md)
- [Vitalik Buterin puts privacy and quantum resistance front and center in Ethereum’s latest roadmap](https://longbridge.com/en/news/295430858.md)
- [Flag Ship Acquisition Targets AI Compute, Aircraft Engines for Cash Flow](https://longbridge.com/en/news/294852890.md)
- [SharpLink CEO Warns Against New Ethereum Network Proposal EIP-8363](https://longbridge.com/en/news/295307613.md)