Stop Asking About TPS: Why Peak TPS Is a Misleading Metric

Every week another blockchain claims 10,000, 50,000, or even 100,000 transactions per second, which raises the obvious question: measured how, exactly?

The TPS arms race has become a running joke among infrastructure engineers because press-release numbers rarely reflect what a system can sustain in production under real workloads across a distributed environment.

When enterprises and financial institutions evaluate a blockchain for settlement, custody, or tokenized asset infrastructure, they run audits focused on sustained throughput, worst-case latency, fee stability, and deterministic finality.

The TPS Number Is Almost Always a Lab Result

Most published blockchain TPS figures are not mainnet numbers. They are theoretical ceilings generated under controlled testnet conditions using simple token transfers on a single node or small validator cluster, without real congestion or state complexity.

For instance, Solana has at various points claimed theoretical throughput upward of 65,000 TPS. Independent measurements of user transactions on mainnet typically range from hundreds to low thousands of TPS depending on methodology and whether validator vote transactions are included.

Similar gaps between benchmark results and sustained mainnet performance appear repeatedly across high-throughput chains. EOS, for example, once claimed 4,000 TPS, while independent testers measured roughly 50 under realistic conditions.

The incentive to publish the highest possible number before launch is understandable, but users have repeatedly been burned by taking those numbers at face value.

Four Metrics To Care About

Sustained Throughput

The real question is whether a network can sustain its claimed rate for hours, not milliseconds, under realistic mixed workloads including complex smart contracts as the state database grows. Very high TPS figures almost always come from benchmark tests run under ideal conditions. Those figures should be treated as upper bounds, not guaranteed production performance.

Worst-Case Latency

Institutional infrastructure cares far more about 99th percentile latency than about averages. A single transaction that takes 30 seconds to finalize because the network spiked can cascade into reconciliation failures, compliance flags, and client escalations. High throughput and low latency are related but distinct properties of a blockchain system.

Fee Stability

Volatile transaction fees are incompatible with institutional operations. A treasury team modeling settlement costs for a billion-dollar tokenization program needs to understand fee behavior at both 50,000 and 500,000 daily transactions. A fee model that is cheap at low volume but spikes unpredictably under congestion creates financial and operational risk. Networks that optimize purely for high TPS often sacrifice fee predictability in the process.

Deterministic Finality

Probabilistic finality, where a transaction becomes "probably irreversible" after some number of block confirmations, is not always acceptable for regulated financial applications. Many regulated financial applications require certainty about when a transaction is settled.

Framework for Evaluating Blockchain Performance

If a vendor cannot produce reproducible numbers for worst-case latency and fee stability under load, the headline TPS figure tells you nothing useful.

The Benchmark Is Not The Only Scoreboard

Performance benchmarks exist to establish trust, not to produce rankings.

An institution deciding to build settlement infrastructure on a blockchain is making a multi-year commitment. They need confidence the performance they model today will hold six months into production during peak usage and with a larger transaction history.

That is a very different bar than "we hit 100,000 TPS in a 10-second test window last Tuesday".

Institutional use cases specifically require predictability and uptime above all else. The blockchain infra that wins institutional adoption is the most consistent under real production conditions.

At Altius, we focus on upgrading the performance of existing chains without requiring them to rewrite their infrastructure. The benchmarks we publish are designed to be audited. The best time to demand better benchmarks was at the start. The second best time is now.