Ethereum founder Vitalik posted on the X platform stating that increasing Ethereum's bandwidth is safer than reducing latency. With the support of PeerDAS and ZKP, we already clearly understand how to scale, and in terms of potential, compared to the current state, theoretically, it can achieve thousands of times expansion. Relevant parameters will become more favorable before and after scaling. Physical laws do not present any obstacles to the coexistence of "extreme scale + decentralization."

Reducing latency is completely different. We are fundamentally limited by the speed of light, and in addition, we face the following practical constraints:

Nodes (especially validators) need to be supported globally to operate in rural areas, homes, or commercial environments, not just data centers

Nodes (especially proposers and validators) need to be provided with censorship resistance and anonymity

Running nodes in geographically non-highly concentrated locations must not only be "feasible" but also economically sustainable. If staking outside New York results in a 10% decrease in returns, more and more people will choose to stake only in New York in the long term

Ethereum itself must pass the "walkaway test," so we cannot build a blockchain that relies on continuous social coordination to maintain decentralization. Economic incentives cannot bear all the responsibility but must bear most of it.

Of course, without making trade-offs, we can still significantly reduce latency on the existing foundation, including:

P2P network improvements (especially erasure codes) can reduce message propagation time without requiring individual nodes to increase bandwidth

An availability chain with fewer nodes per slot (for example, 512 nodes instead of 30,000) can eliminate aggregation steps, allowing the entire critical path to be completed within a subnet

These improvements are expected to bring a 3–6 fold increase. Therefore, I believe moderately reducing latency to the 2–4 second level is entirely realistic.

But Ethereum is not a global video game server; it is the global heartbeat.

If you need to build applications faster than the "heartbeat," they must include off-chain components. This is also one of the important reasons why L2 will still exist long-term even if Ethereum achieves large-scale scaling (other reasons include virtual machine customization and demand for even more extreme scale).

In the longer term, AI will inevitably give rise to applications "faster than the heartbeat." If an AI thinks 1000 times faster than humans, then for it, the "subjective speed of light" is only 300 km/s. This means it can achieve near-instant communication within a city but cannot cover longer distances. Therefore, AI-oriented applications will inevitably emerge, requiring "city-level blockchains," or even single-building-level chains — these can only be L2s.

Conversely, it is unrealistic to pay huge costs to run staking nodes on Mars; even Bitcoin does not pursue this. Ultimately, Ethereum belongs to Earth (Terra), and its L2s will simultaneously serve ultra-localized needs within cities and planet-scale ultra-large-scale