Ethereum co-founder Vitalik Buterin said in a public talk at the 12th Global Blockchain Summit in Shanghai that the industry is moving beyond a model centered on blockchain alone and toward a broader technical stack built from blockchains, zero-knowledge proofs, programmable cryptography and other advanced tools.
Looking back on the past 15 years, Buterin said the surface-level conversation may still sound familiar, with blockchain, crypto and cryptocurrencies remaining at the center, but the underlying technology and the industry’s understanding of its goals have changed in major ways. What users can get today, and what they have to pay for those features, is very different from what people imagined 10 or 15 years ago, he said.
From blockchain as the main tool to a wider stack
Buterin said Bitcoin was born in 2009 and blockchain-related events in Shanghai have been running for more than 10 years. Fifteen years ago, blockchain was the main tool available to the industry. Bitcoin combined blockchain, digital signatures and proof-of-work. Ethereum added smart contracts and later switched to proof-of-stake in 2022.
In his account, the first major property blockchain offered was open participation. Anyone could join. Blockchain systems were global, reliable, secure and neutral, and there was no centralized actor that could interfere with the chain, stop an application from running or force it to produce the wrong result.
That openness came with a tradeoff. Public blockchains are public. Buterin cited a line often used by Zcash founder Zooko: public blockchains are like 「your bank account’s Twitter」.
He said Satoshi Nakamoto’s 2009 white paper offered a basic answer to privacy: real-world names would not be directly attached to transactions. At the time, that was seen as fairly strong privacy protection. Today, with powerful data analytics and AI, he said, that assumption is clearly no longer enough.
At the same time, users want blockchains to do more. They want higher performance, and they are no longer interested only in payments. Even payments themselves are expected to support richer features. For that reason, he said, the technology has to go further.
Zero-knowledge proofs are moving into live protocol design
Buterin described the rise of zero-knowledge proofs as one of the most important technical shifts of the past decade. He pointed to SNARKs and to STARKs, which he described as the quantum-resistant version of SNARKs.
These systems make it possible for one party to perform a computation and generate a proof, while everyone else verifies the proof instead of rerunning the full computation. Verification is much cheaper than recomputing everything, and the process does not require all of the underlying information to be made public.
For blockchain systems, he said, verification is one of the central technical problems. The goal is to build a system that anyone can participate in while still allowing everyone to be confident that the rules are being followed. The original approach was to have everyone rerun every computation. That sacrificed scalability and privacy at the same time. SNARKs and STARKs change both.
Buterin said he has been talking about this direction for more than a decade, and each year it gets closer to real deployment. What changed this year, in his view, is that Ethereum’s plans to adopt these technologies are no longer part of a distant roadmap. They are turning into concrete Ethereum Improvement Proposals, or EIPs.
He cited EIP-8288 as one example. He said the proposal is intended for H-Star, a hard fork that could happen in about two years, though he added that AI could potentially speed things up. He also referred to the Straw Map on strawmap.org and said that after the next two hard forks, everything in that plan will make broad use of STARKs. He called that a major shift.
FHE, iO and stronger programmable cryptography
Buterin also pointed to more powerful cryptographic tools that are starting to look more practical, including fully homomorphic encryption, or FHE, and obfuscation, or iO. For a long time, he said, many people assumed these technologies were far from real-world use. Now, some versions may become viable faster than many expected.
Those tools open up new design space. One example he gave was parallelization in blockchain systems, including parallel work that can begin before transactions are packaged into a block.
FOCIL and EIP-8288 push more work off-chain
In the talk, Buterin said FOCIL is an EIP planned for H-Star and represents Ethereum’s first mechanism for multi-participant block building. Instead of having a block created by a single participant, the process would involve 16 inclusion list creators, with a builder assembling the final block.
The goal is to provide a stronger guarantee that if a user sends a transaction, it can be included in the next slot without any single participant being able to delay it on their own.
He returned to EIP-8288 as another example. The basic idea, he said, is that some proofs are extremely large and expensive to process, but there is no need to place all of them directly into a block. Instead, they can be verified off-chain through a large parallel process.
He located that process in the mempool, the space between a user sending a transaction and that transaction actually landing on-chain. In that stage, many complex objects can be replaced by a proof that attests to all of them.
What the system protects is expanding from assets to information
For Buterin, the deeper significance of these technical changes is that they allow the industry to rethink what these systems are meant to protect.
Bitcoin’s answer, he said, was a digital asset: bitcoin as money. Ethereum extended that model by allowing users to issue their own digital assets and have those assets protected as well.
With modern cryptography, the protected object is no longer limited to digital assets. It also includes information.
He framed the shift this way: in Bitcoin and Ethereum, cryptography and code mainly answered the question of who can send what. Which asset belongs to whom? Who can send which asset, and when? The industry is now entering an era of programmable cryptography in which code starts to decide who can see what. The system is no longer only about who may transfer assets. It is also about who has the right to view information, and that logic is becoming programmable.
Blockchain as a pipeline rather than a single chain
Another major change, Buterin said, is where computation happens. Fifteen years ago, blockchain was just blockchain. Over the next five years, and to a large extent already today, the perspective is shifting from a single chain to an entire pipeline, a large cryptographic network.
In that full process, users first perform substantial computation locally. A lot of work can be done before any data leaves the user’s device. Transactions then enter the mempool. There can be multiple kinds of mempools with different privacy properties. After that, transactions are aggregated. Block construction can involve different kinds of participants, and only then do transactions actually reach the chain.
He contrasted execution models as well. In Bitcoin, every node has to process the same batch of transactions again. In Ethereum, one proof can be verified many times.
The tradeoff shifts from scalability to scalable generality
Buterin said both Bitcoin and Ethereum made major sacrifices in scalability and latency. He compared those systems with centralized infrastructure, saying a centralized system can sometimes reach 100 milliseconds of latency, and even a global system might manage 300 milliseconds. Bitcoin’s latency is about 600 seconds, or 10 minutes. Ethereum currently produces a block every 12 seconds, and finality takes about 16 minutes.
He said Ethereum’s future block time could fall to 4 to 8 seconds, while finality could drop to 8 to 32 seconds.
On scalability, his argument was that future blockchain architecture does not necessarily need to give up nearly as much scalability as before. The real tradeoff will not be scalability itself, but what he called scalable generality.
If a system imposes no constraints at all and offers no guarantees about how computation behaves, he said, then extremely high computational scalability is not possible. But if computation can be split into parts, executed separately, verified separately and processed before anything reaches the chain, then very high scalability becomes achievable.
Viewed as a full system that includes L1, L2 and the mempool, blockchain architecture gains a complex middle layer made up of mempools, L2s and other systems. Users, or their agents, can decide which parts of that middle layer to route through, and much of the scaling capacity can live there.
He summarized the shift in simple terms: the conversation used to be about a ledger. Now it is about cryptographically secure computation. In his words, the industry is bringing computation itself into the cryptographic system and building a true 「cryptographic world computer」, where transactions are processed through multiple cryptographic scaling mechanisms.
In that model, the chain still exists, but its role is reduced as much as possible. The chain no longer has to do everything. The amount of business logic that every node must process in sequence can shrink sharply because most of the work can happen off-chain.
AI is accelerating development, verification and understanding
Buterin said these changes are closely tied to AI. He argued that AI has made a major contribution to the progress now underway. He even joked that his slides looked better because AI helped improve them, specifically thanking Qwen for making them more presentable.
More seriously, he said AI is speeding up every part of the workflow. Before AI, zero-knowledge proofs already existed, but developers could only build relatively simple cryptographic applications. One reason was that once code grew beyond roughly 100 lines, or even 1,000 lines, bugs became almost inevitable.
With AI, developers can use formal verification along with different kinds of redundant testing, specifications and analysis, making it possible to keep larger and more complex systems secure. In the AI era, he said, more work can be moved into cryptographic systems.
He also discussed overhead. In the past, the algorithms the industry used came with very high overhead. Now, 100x overhead or lower is often achievable. For ZK-SNARKs, the overhead for AI inference is already below 10x. Some types of hashing are also below 10x. FHE is still around 250x, but he said that figure is falling very quickly.
Formal verification as a response to L2 security limits
Buterin said one major practical bottleneck, and one reason L2s have not yet fully realized their potential, is that there used to be a ceiling on the security guarantees available to them.
The older approach required redundancy, multiple proof systems and security councils. The newer answer, he said, is formal verification.
If needed, developers can even use multiple specifications and one implementation. Different people, different AI agents and different teams can describe what the system is doing in different ways, and mathematical proofs can then automatically confirm whether they are expressing the same thing.
If those descriptions produce different results, the base set of facts that everyone needs to inspect becomes much smaller, making it easier to determine which side is correct.
Ethereum’s roadmap after H-Star
Buterin said the Ethereum ecosystem is making substantial progress at both the protocol and application layers, and he encouraged developers to study the protocol roadmap.
If you look closely, he said, every hard fork after H-Star, meaning all hard forks starting roughly a year from now, will involve advanced cryptographic features: putting everything into STARKs, putting everything into the virtual machine, aggregating STARKs, formally verifying everything and adopting quantum-resistant signature algorithms.
Taken together, those technologies are turning Ethereum from a simple blockchain into a larger network with multiple layers.
A call to developers to revisit applications
Buterin said one benefit of AI is not only that it helps the industry build these systems faster, and not only that it can help build them more safely if safety is treated as a serious goal. It also makes the technology easier for more people to understand.
He urged developers and researchers to understand the different pieces now being built, understand the properties of each component and think about whether there is a specific area where they want to contribute.
The barrier to participating in frontier technology and ecosystem work is falling quickly, he said. People who are willing to learn, think and invest effort into research and construction have a growing chance to take part directly. He added that there are already public competitions focused on improving algorithms and formally verifying those improvements.
He also said this is a good time to rethink existing applications and build new ones. Many Ethereum applications built five, 10 or 15 years ago were not designed with today’s new capabilities in mind, such as programmable privacy or the possibility of giving strict security guarantees even to very complex code.
Those things are now starting to become possible, he said, and he encouraged developers to return to application building and connect Ethereum, other decentralized networks and other parts of the real world.