Ethereum was not created to make finance efficient or practical applications. It was designed to liberate people.
This phrase from the No Trust Manifesto attracted criticism when it was published, and Vitalik Buterin repeated it on January 5.
The argument: Ethereum’s mission fundamentally differs from the efficiency game in which DeFi protocols compete. The goal is not to achieve 4.5% efficiency versus 5.3%, or to reduce latency from 473 milliseconds to 368, or to reduce registration from three clicks to one.
Ethereum’s game is resilience: avoiding total losses when infrastructure collapses, governments become hostile, or developers disappear. Resilience means maintaining latency of 2000 milliseconds to 2000 milliseconds even when Cloudflare fails, sponsors declare bankruptcy, or users are downgraded.
Resilience means remaining a leading player regardless of geography or politics.
This is important because Ethereum anchors nearly $74 billion of smart contract value in its Layer 1 alone and over 65% of real-world tokenized assets.
Yet the system designed to become the world computer relies on a surprisingly fragile stack of centralized choke points.
The consensus protocol continued to finalize blocks, but the RPC provider’s outdated client caused the exchanges to crash. The blockchain continued to function, but the CDN went dark, taking half the ecosystem offline.
Disaster prevention rather than yield optimization
A recent report quantifies the stakes: infrastructure failures produce volatility shocks 5.7 times greater than regulatory announcements on major crypto assets. The extreme risk of complete loss of access, permanent blocking of funds and network shutdown matters more than additional returns.
A protocol offering a 5.3% yield is worthless if a configuration error can destroy the infrastructure.
Vitalik Buterin’s framing captures this. Resilience isn’t about speed when everything works, but rather whether or not your application works when infrastructure providers disappear or when hosting platforms drop users.
The 2,000 millisecond latency provided by Ethereum may be slower than Web2, but it continues to run even when Web2 systems shut down completely.
Yet Ethereum’s promise of resilience faces practical tests.
As of November 2020, Infura, the default RPC provider for MetaMask and most DeFi applications, was running an outdated Geth client that deviated from the canonical chain.
Exchanges halted Ethereum withdrawals, explorers showed conflicting states, and MakerDAO and Uniswap went bankrupt for users.
Although the bug itself has been fixed and progress is being made on alternative implementations of RPC, centralization remains the norm. It’s just less about Infura only and more about “small cartels”.
The protocol worked, but the attachment points failed.
In November 2025, a Cloudflare configuration error removed approximately 20% of web traffic, including Arbiscan, DefiLlama, and several exchange and DeFi front-ends. Ethereum continued to process blocks. Users could not access it.
During the 2024 registration craze, Arbitrum’s unique sequencer stalled for 78 minutes. No transactions processed, no batches published on Ethereum.
Arbitrum, Optimism, Base and zkSync all currently rely on single, centralized sequencers. The decentralized base layer worked fine, but the centralized infrastructure prevented users from benefiting from it.
| Layer | Current addiction | Fragility measurement | Resilient alternative |
|---|---|---|---|
| Access / RPC | Infura, Alchemy, QuickNode; MetaMask defaults to Infura | ~90% of Web3 application traffic; Infura outage in November 2020 halted ETH withdrawals and broke MetaMask, MakerDAO and Uniswap | Multiple RPC providers, local thin clients, stateless clients as standard; RPC diversity as a user-facing feature |
| Relay / Manufacturer | MEV-Boost relay (Ultra Sound, Titan, bloXroute) mediating >90% of blocks | Four relays control >85% of the proposals; Titan, Beaverbuild, Rsync produce >80% of building blocks | No more relays by separate entities; neutrality of the relay; Dedicated PBS where relay failures cannot block block space |
| L2 sequencing | Unique sequencers (Arbitrum Foundation, Optimism Foundation, Coinbase for Base) | Arbitrum: 78 min stoppage time; Base captures 70.9% of L2’s profits, Arbitrum 14.9%, Optimism 5.4% | Decentralized sequencer games or L1 fallback; force inclusion when the sequencer censors; track % L2 TVL under single control |
| DNS/CDN | Cloudflare for DNS, TLS, and dApp caching | Cloudflare ~20% of the global web; November 2025 outage took Arbiscan, DefiLlama and Exchange/DeFi front-ends out of service | IPFS/Arweave with ENS fallback solutions; multi-CDN; wallets calling contracts without web interface |
| Basic protocol | Ethereum consensus (Lighthouse 52.65%, Prysm 17.66%); execution (Geth ~41%, Nethermind 38%) | The September 2025 Reth bug blocked 5.4% of nodes; diversity prevented broader impact | No customers > 33% share; home staking; minimize correlated failures; easy thin/stateless client verification |
The core protocol demonstrates true resilience, with multiple clients, hundreds of thousands of validators, and proof-of-stake that spreads risk across diverse codebases.
When Reth encountered a bug in September 2025, it blocked 5.4% of nodes, but network continuity was maintained thanks to Geth, Nethermind, and Besu. The diversity of customers worked.
The problem is concentrated above: RPC access, relays, sequencers and web front-ends introduce dependencies that disable user access even when the base layer is running.
This is where Ethereum’s resilience breaks down: not in the cryptography or consensus, but in the scaffolding that connects users to the protocol.
Centralized sequencers as economic choke points
Layer 2 sequencers concentrate both control and profit. Base captured over 50% of all cumulative profits consistently throughout 2025, followed by Arbitrum.
The Arbitrum sequencer is managed by the Arbitrum Foundation, the Optimism sequencer by the Optimism Foundation, the Base sequencer by Coinbase and the zkSync sequencer is centralized.
As a result, over 80% of fees captured by Ethereum Layer 2 in 2025 were directed to blockchains with centralized sequencers.
The technical route exists: networks of shared sequencers like Espresso, or based rollups that send the sequencing back to Ethereum validators. Astria attempted similar designs but closed in 2025.
The difference is not technical, but economic. Centralized sequencers offer better UX and generate substantial revenue. Resilience requires accepting that a sequencer producing slightly slower confirmations, but impossible to stop by a single operator, outperforms millisecond improvements with single-point control.
RPC and CDN dependencies
MetaMask defaults to Infura. Reports indicate that most Web3 applications use Infura, Alchemy or QuickNode.
The Infura incident of November 2020 demonstrated the consequence: resiliency at the protocol level no longer matters when the access layer fails.
The Cloudflare outage in November 2025 revealed the extent to which “decentralized finance” depends on a company’s CDN. Ethereum was processing blocks normally, but users could not access the frontends, explorers, or dashboards.
Resilient alternatives include wallets that default to multiple RPCs, local thin clients, distributed storage on IPFS or Arweave, ENS addressing, and multi-CDN deployments.
However, this imposes costs, such as increased complexity, greater bandwidth requirements, and more complex management.
Most projects choose convenience, so the trade-off in efficiency is important. Ethereum’s base layer provides survival properties, while the ecosystem primarily wraps them in dependencies that reintroduce any fragility.
The real compromise
Ethereum’s value proposition, as Buterin presents it, is not faster, cheaper, or more convenient. It works when everything else fails.
This requires infrastructure choices favoring survival over optimization: multiple client implementations when one is technically superior, various RPC providers when one offers better latency, decentralized sequencers when centralized operators provide faster confirmations, and distributed front-ends when centralized hosting is simpler.
The industry did not accept this compromise. Rollups are optimized for UX and accept the risk of a single sequencer. Applications default to convenient RPCs and accept concentration risk. Front-ends are deployed on commercial CDNs and are tolerant of single-vendor outages.
The choice: build for the case Cloudflare, Infura and Coinbase continue to operate, or build for the case they don’t.
The Ethereum base layer allows for the second choice. The surrounding ecosystem is mainly the first.
The protocol offers 2,000 milliseconds of latency that persists despite infrastructure outages, deplatforms, and geopolitical disruptions.
Whether someone builds systems that actually exploit this property rather than wrapping it in dependencies that reintroduce all the fragilities that Ethereum was designed to eliminate determines whether resilience becomes real or remains theoretical.
Block space is abundant. Decentralized, permissionless, and resilient block space is not.
