Blockchain Trilemma: Security, Scalability & Decentralization

Introduction

A Bitcoin full node costs an estimated $289 in hardware, and a Solana full node costs $21,478, a 74-fold gap that makes the blockchain trilemma concrete (ARK Invest and Glassnode, September 2026). Ethereum sits between the two at $730, and its layer 2 networks add about 2,310 transactions per second to layer 1's 21. The trilemma says a network can maximize only two of security, scalability and decentralization. Solana processes roughly 1,280 user transactions per second against Bitcoin's 6.8, while their Nakamoto coefficients of 19 and 3 look contradictory until the thresholds are compared. This article defines the three properties, shows how Bitcoin, Ethereum, Solana and modular designs choose, explains how to measure each corner, and sets out why critics call the framing a heuristic.

What Is the Blockchain Trilemma and Where Did the Idea Come From?

The blockchain trilemma is the claim that a network can reach only two of three goals at once: security, scalability and decentralization. Vitalik Buterin introduced it in 2017, and the crypto trilemma still frames the argument over how fast a chain can run before it stops being open.

Three properties defined

Buterin defines each property against one test, a consumer laptop (Buterin, 2021). Scalability means the chain processes more transactions than a single regular node can verify. Decentralization means no trust dependency on a small group of large actors, and no honest-majority assumption about nodes that a laptop cannot join. Security means the chain resists a large share of participating nodes attacking it, ideally 50%, where anything above 25% is acceptable and 5% is not (Buterin, 2021). The definitions are operational. Hardware cost, node count and attack thresholds can all test them, so the later sections measure each property instead of arguing about it.

Where the idea came from

Buterin introduced the scalability trilemma in 2017 (Buterin, 2024) as the problem statement that sharding research set out to solve (Buterin, 2021). The original post offered a heuristic argument, not a proof, and its purpose was to show how hard it is to scale a chain without losing decentralization. Ethereum adopted a rollup-centric roadmap in October 2020, after sharding research solved how to verify data availability at scale (Buterin, 2020; Buterin, 2024). Today the term covers any chain's choice among the three goals, from Bitcoin to layer 2 networks.

Every one of those definitions points back to the same bottleneck: the hardware an ordinary person owns.

Why Does Higher Throughput Cost Decentralization or Security?

Higher throughput raises the cost of running a node, and a chain with fewer nodes needs fewer parties to attack or coerce it (Buterin, 2024). That chain of cause and effect is the scalability vs decentralization trade-off at the centre of the trilemma.

More transactions mean heavier nodes

Buterin's argument is arithmetic. A consumer laptop verifies N transactions per second, and a chain processes k times N. Then either each transaction reaches only 1/k of the nodes, so an attacker needs to corrupt only a few of them to push through a bad transaction, or the nodes become expensive and the chain loses decentralization (Buterin, 2024). Bitcoin shows the constraint directly. Blocks are capped at 4 million weight units, and every full node verifies every transaction. Raising the cap lifts throughput and also lifts storage, bandwidth and processing costs for every participant. A faster connection does not remove the limit either, because nodes need spare capacity for attacks and for everything else the machine runs (Buterin, 2021).

The three easy ways out

Buterin sorts the simple designs into three classes (Buterin, 2021). Traditional blockchains such as Bitcoin keep full verification on every node, so they hold decentralization and security but not scalability. High-throughput chains rely on 10 to 100 nodes ↗ that maintain consensus among themselves, so they hold scalability and security but not decentralization, because users must trust a majority of those nodes. Multi-chain ecosystems spread applications across many chains ↗ and gain decentralization and scalability, yet an attacker needs a majority on a single small chain, so they lose security. Each class gives up exactly one corner.

Real networks blend these archetypes, and the cost of running a node shows where each one landed (ARK Invest and Glassnode, 2026).

How Do Bitcoin, Ethereum, Solana and Modular Chains Each Choose?

Bitcoin chose decentralization and security, Solana chose throughput, and Ethereum chose to scale outward through rollups while keeping its base layer verifiable. An estimated full node costs $289 on Bitcoin, $730 on Ethereum and $21,478 on Solana (ARK Invest and Glassnode, September 2026).

Bitcoin keeps verification cheap

Bitcoin Core runs on a 750 GB disk and 2 GB of RAM (Bitcoin Core, 2026). ARK Invest and Glassnode found about 63% of Bitcoin nodes behind Tor and only about 16% in data centers (ARK Invest and Glassnode, 2026). Throughput pays for that accessibility. The network confirmed 589,829 transactions in roughly 24 hours to 5 November 2026, about 6.8 transactions per second (mempool.space, 5 November 2026). Mining is the concentrated layer. Foundry USA, AntPool and F2Pool produced 601 of 974 blocks, or 61.7%, in the seven days to 5 November 2026 (mempool.space, 5 November 2026).

Ethereum and rollups scale outward

Ethereum asks more of a node than Bitcoin does. EIP-7870 recommends 4 TB of NVMe storage, 32 GB of RAM and 50 Mbps of download bandwidth for a full node (EIP-7870, 2025). Layer 1 processed about 21 transactions per second in the seven days to 4 November 2026, while layer 2 networks tracked by L2BEAT averaged about 2,310 combined (L2BEAT, 5 November 2026). Rollups move execution off layer 1 and post the data back ↗, so throughput grows without raising each node's load. The cost shifts to trust assumptions. L2BEAT, renamed ETHBEAT on 4 November 2026, lists 83 active layer 2s, and six had reached Stage 1 and four Stage 2 (L2BEAT, 5 November 2026).

Solana and modular designs

Solana pushes throughput onto the node ↗. Anza's validator documentation calls for 12 cores and 24 threads or more, 256 GB of RAM or more, and separate NVMe drives for accounts, ledger and snapshots (Anza, 2026). Nearly 100% of Solana validators run in commercial data centers (ARK Invest and Glassnode, 2026). Modular designs split the jobs instead. Celestia handles only ordering and data availability, and its light nodes check availability by sampling small pieces of each block (Celestia docs, 2026). Its token, TIA, sits 97.8% below its 10 February 2024 high (CoinPaprika API, 5 November 2026).

Data current as of November 2026.

Table reflects data as of 5 November 2026.

Hardware cost is the cleanest proxy, but the other measures do not rank the networks in the same order.

How Do You Measure Security, Scalability and Decentralization in Practice?

Each corner of the trilemma has a measurable proxy, and none of them transfers cleanly between chains. Nakamoto coefficients of 3, 3 and 19 look comparable but count different entities at different thresholds, and Solana's headline throughput mixed validator votes with user activity until Alpenglow moved votes off-chain on 21 October 2026, so figures from before and after that date do not compare directly.

Decentralization proxies

Balaji Srinivasan and Leland Lee proposed the Nakamoto coefficient in July 2017 as the minimum number of entities needed to reach 51% of a subsystem's capacity (Srinivasan and Lee, 2017). ARK Invest and Glassnode reported 3 for Bitcoin at a 51% hash-rate threshold, 3 for Ethereum at a 33% staked-ETH threshold and 19 for Solana at 33% of delegated stake (ARK Invest and Glassnode, 2026). The thresholds differ, so the figures do not compare directly, and the original authors note that the result depends on which subsystems are counted (Srinivasan and Lee, 2017). Solana Compass showed 18 on 6 September 2026, and the Solana Foundation recorded 20 using April 2025 data, so every figure needs its source and date (CryptoSlate, 2026).

Throughput, claimed and observed

A transactions-per-second figure needs a definition before it carries meaning. Solana's total rate, about 4,000 to 4,300 transactions per second in samples on 23 September 2026, included validator votes, which were ordinary transactions until Alpenglow moved votes off-chain on 21 October 2026. Non-vote throughput measured 1,400 to 1,800, and that figure still includes failed transactions (Solscanner, 23 September 2026). Since the switch, the total equals the non-vote count, which averaged about 1,277 per second in the hour to 10:48 UTC on 5 November 2026 (Solana RPC, 2026). Bitcoin's 6.8 transactions per second is a measured rate, and Ethereum's layer 1 ran at about 21. Theoretical maxima describe block parameters rather than live traffic (Solscanner, 2026). A rate also says nothing about the cost of the node needed to keep up, which is the other half of the trade-off.

Data current as of November 2026.

The metrics expose the trade-off, and they also expose how much of it depends on definitions.

Is the Blockchain Trilemma a Real Limit or Just a Heuristic?

The trilemma is a heuristic, and its author says so. Buterin writes that it is not a theorem and that the original post carried no mathematical proof (Buterin, 2024). Critics make two points: new techniques may break the trade-off, and the choice of metric can change which chain looks most decentralized.

A heuristic, not a theorem

The original argument showed that breaking the trilemma is hard, not that it is impossible. Its purpose was to show that escaping the trade-off requires thinking outside the box the argument sets out (Buterin, 2024). Academic work has since attempted a formalization. Mssassi and Abou El Kalam published a paper titled "A Formal Proof of the Inherent Trade-Offs Among Decentralization, Security, and Scalability" in Applied Sciences on 24 December 2024 (MDPI, 2024). Buterin separates architecture from software engineering. Chains that claim to solve the trilemma by optimizing node software are misleading in his words, because running their nodes ends up far harder than running an Ethereum node (Buterin, 2024).

Where the framing misleads

Data availability sampling combined with SNARKs does solve the trilemma, in Buterin's assessment, because clients can verify that data is available and computation correct while handling a small share of both (Buterin, 2024). Sampling has a few-of-N trust model, yet it keeps the property that a 51% attack cannot force bad blocks onto the network (Buterin, 2024). Ethereum's Fusaka upgrade activated PeerDAS ↗, a sampling scheme, on 3 December 2025 (Ethereum Foundation, 2025). Metrics can also mislead. ARK Invest and Glassnode rank Bitcoin first and Solana third overall even though Solana's Nakamoto coefficient is the highest of the three (ARK Invest and Glassnode, 2026). Layer 2s move the trade-off rather than remove it, into sequencers, upgrade keys and exit windows (L2BEAT, 2026).

FAQ

?Is the blockchain trilemma the same as the CAP theorem?

No. The CAP theorem, short for consistency, availability and partition tolerance, is a distributed-systems result: a system facing a network partition must choose between consistency and availability. Eric Brewer conjectured it in 2000, and Seth Gilbert and Nancy Lynch proved it in 2002 (Gilbert and Lynch, 2002). The blockchain trilemma is a heuristic about node cost, throughput and attack resistance, and its author says it has no proof (Buterin, 2024). Both frame three goals of which two are reachable, but they cover different properties.

?Does "security" in the trilemma mean protection from hacks and smart contract bugs?

No. In the trilemma, security means the chain resists attack by a large share of its participating nodes, ideally 50% (Buterin, 2021). It concerns consensus: how much stake or hash rate an attacker needs to rewrite or censor the chain. Smart contract bugs, exchange failures and stolen keys sit outside that definition. L2BEAT makes a similar point about its stage ratings, which measure decentralization rather than bug risk (L2BEAT, 2025).

?Does proof of stake solve the blockchain trilemma compared with proof of work?

Not by itself. A consensus mechanism changes who pays for security, but the trilemma argument concerns how much each node must verify (Buterin, 2024). Ethereum switched to proof of stake on 15 September 2022 (OpenSea, 2023) and still scales through rollups and data availability sampling. ARK Invest and Glassnode report a Nakamoto coefficient of 3 for Ethereum at a 33% staked-ETH threshold, the same figure as proof-of-work Bitcoin (ARK Invest and Glassnode, 2026).

?How can an ordinary user check how decentralized a chain really is?

Live trackers publish the figures. Nakaflow calculates the Nakamoto coefficient from entity-level staking data and warns that validators run by one party cannot always be identified. Solana Compass defines it as the fewest validators whose stake reaches 33.4% of voting power (CryptoSlate, 2026). Hosting matters as well: about 20% of Ethereum nodes run on Amazon Web Services (ARK Invest and Glassnode, 2026). Always note which threshold a tracker uses before comparing two chains.

?What risk do holders face when a chain gives up decentralization?

Buterin argues that on a chain where regular users run nodes, a 51% attacker can revert or censor transactions but cannot insert invalid ones, and users can detect the behavior and coordinate a fork. High-throughput chains with few nodes lack that check (Buterin, 2021). Holders therefore depend on a small validator group behaving honestly, and the group's concentration, hosting and software all become points of failure.

?Does using a layer 2 mean giving up Ethereum's security?

It depends on the layer 2. L2BEAT tracked 83 active layer 2 projects on 5 November 2026, with six at Stage 1 and four at Stage 2. Stage 1 requires at least 75% of a security council to block or forge withdrawals and an exit window of at least seven days for other upgrades. Stage 1 projects held about 69% of the roughly $38.8B secured, so most value sits in the more trusted tier (L2BEAT, 5 November 2026).

?Who created the Nakamoto coefficient, and what does it measure?

Balaji Srinivasan and Leland Lee proposed it in "Quantifying Decentralization" on 28 July 2017 (Srinivasan and Lee, 2017). It counts the smallest number of entities needed to reach 51% of a subsystem's capacity, and the "minimum Nakamoto coefficient" takes the lowest value across subsystems. The authors built it from the Gini coefficient and Lorenz curve and stress that the result depends on which subsystems are defined as essential (Srinivasan and Lee, 2017).

?Is the trilemma the reason transaction fees spike when a network is busy?

In part. Capacity is finite on any chain where every node verifies everything, and when demand exceeds it, users bid for space. OpenSea's guide describes gas fees as both validator pay and a tool for reducing demand during periods of high utilization (OpenSea, 2023). Bitcoin's block cap of 4 million weight units is the clearest case, and layer 2s add capacity by moving execution off layer 1 ↗.

References / Sources