Layer-1 Scaling:
Beyond TPS Metrics

Transactions per second is a throughput measure. It tells you how many discrete state transitions a network can process in one second. What it does not tell you is the nature of those transactions, the conditions under which they were counted, the cost imposed on the network by processing them at that speed, or what happens to the system when it actually approaches that limit.

// What "65,000 TPS" actually means in most whitepapers:

test_conditions: {
  transaction_type:   "simple token transfer",  // not contract calls
  validator_count:    4,                          // not 1000+
  network_topology:   "local testnet",            // not global p2p
  state_size:         "minimal",                  // not 200GB+
  adversarial_load:   false,                      // no spam
}

⚠  production_tps ≈ benchmark_tps × 0.03 to 0.15
⚠  smart_contract_tps ≈ benchmark_tps × 0.005 to 0.04

Ethereum mainnet processes roughly 15–30 TPS in production. Its theoretical ceiling, often cited at around 30 TPS, is achieved with simple ETH transfers; complex DeFi interactions involving multiple contract calls consume gas that effectively reduces throughput to single digits for that transaction type. Solana's real-world throughput, widely cited in independent analyses, hovers between 2,000–4,000 TPS — impressive, but a fraction of headline claims, and achieved with a validator hardware requirement that has material implications for decentralization.

For most end-user experiences — and for virtually all cross-chain bridges, exchanges, and payments infrastructure — the relevant question is not how many transactions per second a chain can process, but how long after submission a transaction is irreversible. This is finality: the moment at which a transaction cannot be reorganised away by a future chain state.

Finality types differ fundamentally across consensus mechanisms:

For payment applications, instant BFT finality is the gold standard. For DeFi protocols where liveness is paramount, economic finality with strong validator economics may be preferable. For store-of-value use cases, probabilistic finality backed by extreme hash rate remains the battle-tested choice. No single finality model is universally superior — but TPS tells you nothing about which model a chain uses.

The blockchain trilemma — the proposition that networks can optimise for at most two of security, scalability, and decentralization — is a simplification, but a useful one. What it captures is that decentralization has real costs that manifest as throughput limits, and that teams who claim to have solved the trilemma have almost always done so by redefining one of its terms.

Decentralization in practical terms means: how many independent parties must collude to corrupt the chain? This question has multiple layers.

Node count is an imprecise but instructive proxy. A chain with 10,000 full nodes is meaningfully more decentralized than one with 100 — because geographic distribution, client diversity, and hardware accessibility compound. Ethereum's ~10,000 full nodes, distributed across dozens of countries and multiple client implementations (Geth, Lighthouse, Besu, Nethermind), represent a kind of decentralization that is structurally resistant to nation-state-level attack in ways that a 1,000-validator proof-of-stake chain operating within a handful of jurisdictions is not.

The Nakamoto Coefficient

The Nakamoto Coefficient — the minimum number of entities required to reach a protocol-breaking threshold (typically 33% for liveness, 51% for safety in BFT systems) — is a more precise decentralization metric than node count. It can be calculated across multiple subsystems: mining/staking, clients, geographic distribution, exchange liquidity, and developer activity. A chain with a Nakamoto Coefficient of 3 on any subsystem has a single-digit number of attack targets — regardless of what its TPS marketing claims.

Every transaction on a smart contract platform modifies state: account balances, contract storage, code. State is the cumulative record of every interaction ever made. State growth is the rate at which this record expands — and it represents one of the most underappreciated threats to long-term blockchain viability.

// State size as of 2025 (approx, full archive nodes):
ethereum_archive:  ~2.4 TB   // growing ~1 GB/day
ethereum_pruned:   ~1.1 TB  // manageable, but rising
solana_ledger:     ~90 TB   // extremely high due to throughput
bitcoin_full:      ~650 GB  // slow growth, UTXO model helps

// The problem: state must be held in fast storage (SSD/RAM)
// for validators to process blocks within slot time.
// As state grows, hardware requirements escalate.
// Decentralization shrinks. The trilemma bites.

High-TPS chains face a compounding state growth problem: more transactions per second means more state modifications per second. Solana's ledger history, requiring tens of terabytes even with aggressive pruning, is already accessible only to institutional node operators with significant storage infrastructure. This is not a temporary condition — it is a structural consequence of the design choices that enable the throughput figure.

Solutions under active development — Ethereum's EIP-4444 (history expiry), stateless clients, and Verkle trees — address this at the protocol level by separating history storage from state execution. These represent genuine progress. They also illustrate that TPS without a state growth management strategy is not a long-term scalability solution. It is a debt instrument: borrowed performance, payable in future centralization.

Maximal Extractable Value — formerly Miner Extractable Value — is the profit available to block producers by selectively ordering, including, or excluding transactions within a block. It is, in practical terms, a tax on network users levied by validators, and it scales directly with throughput: more transactions per second means more opportunities for value extraction.

MEV manifests as sandwich attacks on DEX trades, frontrunning of oracle updates, liquidation racing in lending protocols, and arbitrage across fragmented liquidity pools. In competitive MEV environments, users pay inflated gas fees as bots bid for priority inclusion, and naive users with large trades lose value to sandwiching that never appears in their transaction receipt.

Chains that claim to eliminate MEV have typically done so through one of three mechanisms: private mempools (which centralise information advantage with the block producer), deterministic ordering (which eliminates MEV but also eliminates priority fee markets, creating inclusion uncertainty), or application-layer rate limiting (which pushes MEV to other surfaces). None of these are free — they represent different distributions of the same underlying economic force.

A blockchain's security model ultimately rests on an economic proposition: attacking the chain must cost more than any achievable gain. For proof-of-work chains this is the cost of acquiring and operating sufficient hash rate. For proof-of-stake chains it is the cost of acquiring and risking the slashing of sufficient staked value.

Economic security is not static. It scales with the market value of staked or mined assets — meaning that during bear markets, chains with smaller market caps see their security budgets compress. A chain with $5B in staked value at $10 token price has a very different security profile from the same chain at $0.50. TPS figures, needless to say, are unaffected by this dynamic.

Rather than a single headline number, a rigorous Layer-1 evaluation examines performance across at least six dimensions, each representing a distinct set of engineering tradeoffs:

The honest answer to "which Layer-1 is best" is: it depends entirely on what you are building and which dimensions of the evaluation framework you are willing to compromise.

For maximum security and decentralization: Ethereum remains the only chain with a combination of economic security, validator diversity, client pluralism, and geographic distribution that approaches meaningful censorship resistance at nation-state scale. Its throughput limitations are real, but its Layer-2 ecosystem (Arbitrum, Optimism, Base, zkSync) means that application throughput and Ethereum security are no longer mutually exclusive.

For applications requiring fast finality and high application-layer throughput: Solana's engineering is genuinely impressive, and its real-world performance in consumer applications (compressed NFTs, high-frequency trading, consumer payments) is unmatched among decentralised alternatives. The validator concentration and hardware requirements represent known, quantifiable tradeoffs — the question is whether they are acceptable for the specific application.

For sovereign application chains: The Cosmos ecosystem's application-specific chain model — where each application runs its own BFT chain with its own validator set and token economics — offers a different kind of scalability: not more TPS on a shared chain, but the ability to design the chain itself around application requirements. The tradeoffs are security fragmentation and bridge complexity.