TPS is a vanity metric that ignores the real constraint: state bloat. Every transaction permanently increases the chain's state size, raising hardware requirements for nodes and increasing sync times. This is the fundamental scalability trilemma.
the-ethereum-roadmap-merge-surge-verge
Blog
Ethereum Throughput Limits That Matter to CTOs
A cynical breakdown of Ethereum's true scalability constraints beyond marketing TPS. We analyze state growth, data availability costs, and the L2 trilemma through the lens of the Surge roadmap, providing actionable insights for infrastructure decisions.
introduction
THE THROUGHPUT FALLACY
Introduction: The TPS Lie
Ethereum's real throughput bottleneck is not raw TPS but the economic cost and finality delay of state growth.
Layer 2s like Arbitrum and Optimism solve execution, not data availability. Their rollup proofs settle on Ethereum, but the underlying calldata still posts to L1. This creates a hard ceiling on total network throughput.
The real limit is gas. Ethereum's block gas limit dictates the maximum computational and storage work per block. Even with EIP-4844 proto-danksharding, the data bandwidth for rollups is capped by this economic parameter, not theoretical TPS.
Evidence: A full Ethereum archive node requires over 12TB of storage. Arbitrum One's sustained throughput is ~0.1M TPS off-chain, but its on-chain data posting is throttled by L1 gas costs, creating a hard, economically-bound ceiling.
key-trends
ETHEREUM THROUGHPUT
Executive Summary: The Three Real Bottlenecks
Forget theoretical TPS. Real-world scaling is constrained by three physical limits that directly impact your application's cost, speed, and security.
01
The Block Gas Limit: The Fixed-Cost Ceiling
Each Ethereum block has a hard gas limit, capping transaction throughput to ~15-20 million gas per block. This creates a volatile, auction-based fee market where user demand directly competes with your protocol's operations.
- Result: Unpredictable and often prohibitive costs for on-chain settlement.
- Impact: Limits the complexity and frequency of on-chain interactions, pushing logic to L2s or off-chain.
~30 TPS
Base Layer Cap
>1000 Gwei
Peak Gas Price
02
State Growth: The Node Operator Tax
Every new contract and storage slot permanently increases the global state size, imposing a linear cost growth on node hardware requirements. This centralizes the network and makes historical data access slower.
- Result: Rising sync times and hardware costs for node operators, threatening decentralization.
- Solution: State expiry proposals (Verkle Trees, The Purge) and pushing state to specialized networks like EigenDA or Celestia.
1TB+
Full Node Size
Days
Sync Time
03
Data Availability: The Rollup Scaling Anchor
Rollups (Arbitrum, Optimism, zkSync) need to post their transaction data to Ethereum for security. The data availability (DA) layer is now the primary throughput bottleneck, limited by Ethereum's ~80 KB/s calldata bandwidth.
- Result: Rollup throughput is gated by L1 data posting costs and speed.
- Evolution: Adoption of EIP-4844 (blobs) and alternative DA layers from Celestia and Avail to decouple execution from data settlement.
~80 KB/s
Current DA Bandwidth
10-100x
Blob Scaling
deep-dive
THE BOTTLENECK
The State Bloat Problem: Throughput's Hidden Tax
Ethereum's fundamental throughput limit is not gas or TPS, but the unsustainable growth of its global state.
State growth is the hard cap. Every new account or smart contract stored on-chain increases the global state size, which every node must replicate and process. This creates a throughput tax where higher transaction volume accelerates node hardware requirements, centralizing the network.
Scaling solutions export the problem. Layer 2s like Arbitrum and Optimism batch transactions to reduce mainnet load, but their own state still grows. This shifts the bloat burden to their sequencers and provers, creating a second-order centralization risk within the L2 ecosystem.
Statelessness is the only viable path. Proposals like Verkle Trees and EIP-4444 aim to make nodes stateless clients. This architecture separates execution from state storage, allowing validators to process blocks without holding the entire history, which is the prerequisite for sustainable scaling.
ETHEREUM THROUGHPUT BOTTLENECKS
Data Availability: The Cost of Settlement
Quantifying the primary constraints and costs for settling transactions on Ethereum, comparing base layer, rollups, and alternative DA layers.
| Constraint / Metric | Ethereum L1 (Calldata) | Ethereum L2 (Rollup) | Celestia (External DA) |
|---|---|---|---|
Max Theoretical TPS (Data) | ~80 | ~3,000 | ~10,000+ |
Blob Data Cost per MB (Current) | $100 - $500 | $0.10 - $1.00 | $0.01 - $0.10 |
Settlement Finality Time | 12-15 min | ~1 hour (to L1) | ~15 min (DA finality) |
Data Availability Guarantee | Strong (Ethereum Consensus) | Strong (via Ethereum) | Weak (Separate Consensus) |
Protocol Revenue Source | EIP-1559 Burn + Priority Fees | Sequencer Fees + MEV | Data Publishing Fees |
Developer Overhead | None | High (Fraud/Validity Proofs) | Medium (Data Availability Proofs) |
Ecosystem Security Budget | $40B+ (Staked ETH) | Inherited from Ethereum | $1B+ (TIA Staked) |
Censorship Resistance | High (Decentralized Proposers) | Medium (Centralized Sequencer Risk) | High (Decentralized Network) |
counter-argument
THE THROUGHPUT CONSTRAINT
The L2 Trilemma: Cheap, Fast, or Secure – Pick Two
Ethereum's L2 scaling solutions force a fundamental trade-off between cost, speed, and security that defines architectural choice.
The base layer is the bottleneck. Every L2, from Optimism to Arbitrum, must eventually settle its state on Ethereum. This creates a hard ceiling on total system throughput, forcing L2s to compete for scarce block space.
Cheap and Fast sacrifices Security. Validiums like StarkEx and certain zkEVMs post only validity proofs to Ethereum, keeping data off-chain. This slashes costs but introduces a data availability risk, requiring trusted operators.
Secure and Cheap sacrifices Speed. Optimistic Rollups like Arbitrum One post all data on-chain, inheriting Ethereum's security. This guarantees censorship resistance but suffers from a 7-day withdrawal delay and higher finality latency.
Fast and Secure sacrifices Cheap. zkRollups like zkSync Era provide near-instant finality with cryptographic security. Their computational intensity and on-chain data posting, however, maintain higher per-transaction costs than optimistic counterparts.
Evidence: During the 2024 memecoin frenzy, average transaction costs on Arbitrum spiked to $0.40, while Base saw surges above $1. This is the trilemma in action: high demand makes 'cheap' impossible without compromising the other vertices.
takeaways
ETHEREUM THROUGHPUT
CTO Takeaways: Architecting for the Real Limits
Ethereum's mainnet is a secure but constrained settlement layer. Scalability is a design constraint, not a bug. Here's how to build for it.
01
The 30 Gwei Ceiling
User demand collapses above ~30 Gwei. This is your real-world TPS limit, not the theoretical ~15-45 TPS. Architect for predictable, not peak, capacity.\n- Design for Gas Spikes: Batch transactions, use gasless meta-transactions via EIP-4337 Account Abstraction.\n- Cost as a Feature: Use this ceiling to model your L2/L3 economic viability.
~30 Gwei
Demand Cliff
15-45
Theoretical TPS
02
The Blob-Carrying Capacity
Post-Dencun, throughput is defined by blob space, not gas. Each slot has ~0.75 MB of dedicated data. This is the new bottleneck for rollups like Arbitrum, Optimism, and Base.\n- Monitor Blob Prices: They are volatile; design fee markets that absorb spikes.\n- Batch Aggressively: Maximize data compression and proof aggregation to stay cost-effective.
~0.75 MB/slot
Blob Capacity
3-6
Blobs per Slot
03
State Growth is the Final Boss
Throughput is meaningless if state size grows exponentially. Full nodes require ~1 TB+ of storage, threatening decentralization. Your dapp's state access patterns are a public good concern.\n- Use Statelessness Primitives: Design for Verkle Trees and EIP-4444 (history expiry).\n- Offload State: Leverage EigenLayer AVSs or dedicated co-processors for heavy computation.
1 TB+
Node Storage
~12s
Sync Time (weeks)
04
L2s Are Not Created Equal
Throughput claims are marketing. Real limits are: proving time, sequencer latency, and bridge finality. A zkRollup (e.g., zkSync, Starknet) has a ~10 min proof finality delay vs. an Optimistic Rollup's 7-day challenge window.\n- Match Finality to Use Case: Use Polygon zkEVM for fast withdrawals, Arbitrum for general purpose.\n- Decentralize the Sequencer: Relying on a single sequencer is a ~2-12s liveness fault risk.
10 min vs 7 days
Finality Range
2-12s
Sequencer Risk
05
The Interoperability Tax
Cross-chain activity multiplies latency and cost. A simple bridge + swap can take 3-5 minutes and cost $10+. This is a throughput killer for composite apps.\n- Embrace Intents: Use UniswapX, CowSwap, Across for optimized cross-domain routing.\n- Standardize Messaging: Rely on LayerZero, CCIP, or Wormhole but understand their security/trust models.
3-5 min
Cross-Chain Latency
$10+
Typical Cost
06
Throughput is a Security Budget
Every TPS costs security. Ethereum's ~$40B staked secures ~30 TPS. A comparable Solana validator requires ~$100k+ hardware. Your chain's security budget dictates its maximum viable throughput.\n- Quantify Security Spend: Compare cost-of-corruption vs. cost-of-attack.\n- Hybrid Models: Use Ethereum for finality, high-throughput systems (Monad, Sei) for execution.
$40B
ETH Security Budget
$100k+
Validator Hardware
ENQUIRY
Get In Touch
today.
Our experts will offer a free quote and a 30min call to discuss your project.
NDA Protected
Confidentiality24h Response
Time to ValueDirectly to Engineering Team
No Sales Fluff10+
Protocols Shipped
$20M+
TVL Overall