Blockchain projects routinely publish throughput benchmarks that bear little resemblance to what happens when the network actually runs. A chain might claim 100,000 transactions per second under lab conditions, then choke at a fraction of that load once validators sync state, storage fills, or network latency kicks in.
The gap exists because standard TPS tests measure only the narrow task of pushing transactions through consensus. They ignore the machinery running underneath. Validators must store full or partial ledger state. Each new block requires I/O operations—reads and writes—that pile up fast. Nodes need to sync with peers, a process that slows dramatically as chain history grows. Genesis syncing, pruning strategies, and network topology all introduce friction that lab tests skip.
NewsData identifies storage bandwidth and input/output operations per second (IOPS) as the primary culprits. A validator hitting its disk ceiling stops accepting new blocks, regardless of what the protocol theoretically permits. Similarly, syncing delays compound when networks are geographically dispersed or when nodes run on consumer hardware. A chain that performs flawlessly in a controlled test with 10 nodes in the same data center will stall when hundreds of validators sprawl across continents.
Why this matters now
Developers and infrastructure providers need to stop anchoring decisions to headline TPS. A chain claiming 1 million TPS might sustain 10,000 TPS on live networks. That gap isn't academic—it shapes whether a builder can actually scale an application, how much decentralization a network can afford, and what validator hardware costs really look like.
Traders and users face a subtler risk. Throughput claims influence which chains attract capital and developer attention. If a project's actual sustained throughput is orders of magnitude lower than its marketing material suggests, adoption timelines slip, and the economic assumptions that priced in rapid growth evaporate.
The hidden constraints
Storage is the blunt force. Blockchains that don't prune aggressively accumulate state at a rate that quickly exhausts even large disks. Pruning itself takes time and compute. IOPS ceilings bite next: validators can't execute more transactions than their disks can write per second. A high-performance SSD might handle 10,000 to 100,000 IOPS. Add network latency—especially if validators must wait for state from geographically distant peers—and effective throughput drops further.
Testing labs typically sidestep these constraints by running validators on beefy hardware in low-latency environments and truncating the test to a few blocks. Real networks have diverse validator hardware, longer history, and unpredictable latency. The performance delta is often dramatic.
Evaluating a blockchain's actual throughput requires stress tests on real network conditions, not isolated benchmarks. Look for published data on sustained throughput (not peak), average validator hardware specs, and observed latency under load. If a project won't share those details, its TPS number is marketing, not engineering.