[Intermediate] — the constraint every chain design is navigating, and how to read a “100,000 TPS” claim.

The claim

Decentralisation, security, scalability — pick two.

  • Decentralisation — many independent participants can verify the chain on affordable hardware
  • Security — attacking it costs more than it yields
  • Scalability — high throughput at low cost

It’s a heuristic rather than a theorem, but it captures a real constraint, and it’s the single most useful lens for evaluating chain claims.

Why the tension is real

Every node executes every transaction. That’s what makes the chain verifiable without trust — and it means throughput is capped by what the weakest participating node can handle.

To go faster you must raise the requirements: more CPU to execute, more bandwidth to receive blocks, more disk to store state. Raise them enough and home nodes drop out. What remains is a fast chain verified by a handful of data centres — which is a database with extra latency.

raise the gas limit
   → bigger blocks
      → more bandwidth and disk needed to keep up
         → fewer people can run a node
            → fewer independent verifiers
               → decentralisation falls

Note that bandwidth, not CPU, is usually the binding constraint — which is unintuitive, and it’s why “hardware is getting faster” doesn’t dissolve the problem.

How each design chooses

ChainTrade
BitcoinMaximum decentralisation and security. ~7 TPS. Deliberate, and the 2015–17 block-size war was fought precisely over refusing to trade it
Ethereum L1Decentralisation and security first; scaling delegated to L2s. ~15–100 TPS
SolanaScalability first. ~3,000+ real TPS, at validator hardware requirements that put a node in a data centre. Has halted several times
BNB ChainScalability, with ~40 validators. Fast, cheap, and decentralised mainly as a claim
RollupsAttempt to escape the trilemma rather than trade within it — see below

Reading TPS claims

Almost every headline TPS figure is theoretical maximum under ideal conditions, and the gap to reality is usually an order of magnitude. Questions that deflate most of them:

  1. How many validators, and what hardware do they need? 21 validators on 64-core servers is a distributed database
  2. Is that sustained or peak? Benchmarks measure bursts; state growth is what bites
  3. What are the transactions? Simple transfers are cheap; contract calls are not. Many benchmarks measure transfers and quote the number for everything
  4. Where does state live in five years? A chain doing 50,000 TPS accumulates state faster than consumer disks grow. Nobody can run a node eventually is a slow failure mode that benchmarks don’t show
  5. Has it halted? Liveness under real load is the actual test, and several high-TPS chains have failed it publicly

The rollup escape

The insight behind Ethereum’s strategy: the trilemma binds because every node re-executes everything. So don’t re-execute — verify a proof instead.

Execution happens off-chain, on a rollup, with one sequencer

    ↓ posts compressed transaction data + a validity proof (or a fraud window)

L1 verifies the PROOF, not the execution

Verifying a zk proof is roughly constant-cost regardless of how much computation it attests to. So L1 can secure thousands of L2 transactions for the cost of verifying one proof. The rollup can use fast hardware and one sequencer, because its correctness is guaranteed by the proof rather than by its own decentralisation.

What’s genuinely escaped and what isn’t:

  • Escaped: execution throughput. Real, and it works
  • Not escaped: data availability. The transaction data must be published somewhere anyone can read, or nobody can reconstruct state or prove fraud. Publishing that data is now the binding constraint — which is exactly what EIP-4844 blobs addressed, and why L2 fees fell an order of magnitude in 2024
  • Not escaped: sequencer centralisation. Most rollups run one sequencer, which can censor and reorder → rollups

Modular vs monolithic

The framing this produced:

  • Monolithic (Solana, Bitcoin) — one chain does execution, settlement, consensus and data availability. Simpler, tightly optimised, and bounded by the trilemma
  • Modular (Ethereum + rollups + Celestia) — each layer specialises. More scalable, more complex, and the seams between layers are where bridges live — and bridges are where the largest losses in the field have occurred → bridges

Neither has won. Solana’s monolithic bet is the strongest counter-argument to the modular thesis, and it is doing well enough to keep the question open.

Key insight

The trilemma is really a statement about verification cost. Decentralisation means many people can independently check the chain, and that caps throughput at the weakest checker. Rollups don’t repeal it — they change what gets checked from execution to a proof, which is a genuine escape for computation and not for data availability. Any scaling claim should be read by asking: what is a node still required to verify, and on what hardware?

Source: [reference] — Aug 2026.

↑ Back to Beyond Ethereum