- 1M TPS achieved: Used real NEAR core code in a realistic, publicly verifiable benchmark.
- Powered by Google Compute Engines C4D machines based on the latest 5th Gen AMD EPYC processors: Demonstrated on cost-effective, consumer-grade hardware to support a decentralized validator network.
- Massive headroom for growth: Showcases NEAR’s ability to scale horizontally for high-load dApps and the onchain AI economy.
NEAR Protocol has reached a significant new performance milestone of 1 million transactions per second (TPS) in a publicly verifiable benchmark using real code, realistic workloads, and cost-effective and performant Google Compute Engine C4D machines across 70 shards.
This result verifies NEAR’s long-standing architectural thesis that sharding is the path to true scalability for the blockchain industry. With Nightshade 2.0, stateless validation, and sharded contract execution via Deterministic Accounts, NEAR is positioned to scale as application demand increases, without compromising decentralization or cost.
Today’s milestone ensures scalable infrastructure for NEAR Intents and opens the door for high-load dApps and the emerging onchain AI economy to run at global scale.
The Road to 1M TPS: NEAR’s Sharding Vision
Most blockchains still rely on monolithic architectures that hit throughput ceilings under real user load. NEAR was designed differently: a sharded L1 where execution, storage, and state are distributed across parallel shards. NEAR’s sharded architecture enables:
- Horizontal scaling so TPS increases linearly as shards are added.
- Low, predictable costs even as usage rises.
- Decentralization through low requirements for validator hardware.
- High-load, real-time applications, including AI agents making rapid onchain decisions.
Last year, NEAR launched the Nightshade 2.0 sharding upgrade on mainnet, introducing stateless validation and greatly increasing NEAR protocol’s efficiency and scalability. Since then, NEAR has scaled to nine shards on mainnet and achieved 600 millisecond blocks and 1.2 second finality. These protocol-level innovations have helped solidify NEAR as the purpose-built blockchain for high-volume cross-chain DeFi and onchain agentic commerce.
Methodology: Designing a Realistic, Verifiable Benchmark
The engineering team set strict requirements to ensure the benchmark was credible, reproducible, and aligned with real-world mainnet environments. Benchmark requirements:
- Real NEAR core code from master branch (or about to be merged).
- Hardware that validators can actually afford: under ~$1k/mo on cloud machines.
- Reasonable shard count: up to 80, demonstrating horizontal scaling.
- Publicly verifiable: anyone can run the same commit and reproduce results.
- Open dashboard with detailed TPS data.
This benchmark was not a synthetic lab demo but rather followed principles similar to production environments and validator constraints to demonstrate NEAR’s throughput in realistic conditions.
Benchmark Setup: 70 Shards, Consumer-Grade Hardware
The benchmark used realistic modeling of L1 user behavior at scale. Key setup parameters:
- 1M total accounts per shard
- Native token transfers only. Contract execution intentionally excluded.
- Account activity followed Zipf’s law and uniform cross-shard traffic, a realistic distribution of activity concentration.
- Transactions injected directly into mempool. RPCs excluded as they are off-chain infra.
- No block gas limit to measure raw throughput potential.
- Three Google Cloud regions for network distribution.
- Machine type: Google’s new c4d-highmem-16 (8-core CPUs) with 200 GB hyperdisk-balanced (440 MB/s, 4200 IOPS) as a boot disk and 400 GB hyperdisk-balanced (1200 MB/s, 20000 IOPS) as an attached disk, total cost ~$900/month in us-central1 (Iowa) at the time of writing.
Google’s new C4D machine types are designed to deliver peak, consistent performance while optimizing costs. These instances leveraged Google’s Titanium architecture, which delivers improved overall resource efficiency, latency, and security benefits. We saw that Google’s C4D VMs combined with Hyperdisk block storage, delivered 45% performance uplift over the prior generation instances that we were using previously, while remaining within our target budget.
Results: 1,000,000+ TPS
The system peaked over 1,000,000 TPS across 70 shards during the benchmark window, with stable block production and predictable throughput. This run is publicly viewable on the shared Grafana dashboard below and can be reproduced by anyone following the provided scripts.
This benchmark used the new Google Cloud C4D VM instance hardware to enable rapid iteration. On bare-metal deployments (used by a majority of mainnet validators), performance would likely be even higher.
Key Optimizations Across the Stack
After profiling and analyzing a series of runs, the team implemented several optimizations across NEAR’s execution and consensus layers to improve flow. Engineering enhancements included:
- Execution layer optimization
- Signature verification improvements
- Trie and database layer rewrites/tuning
- Optimistic execution pathways
- Consensus and stateless witness optimizations
- Network interaction tuning
The team also built a custom trace analysis tool to inspect intra-block timings and chunk production flows, which helped identify bottlenecks and critical paths.
Deliverables: Full Transparency and Reproducibility
To ensure the community and industry can verify these results, we are releasing the following deliverables:
- Open-source benchmark scripts
- Public Grafana dashboards with 1M TPS runs: Run 1, Run 2, Run 3
- Commit hash of the benchmarked code
These materials enable researchers, auditors, validators, and the wider blockchain ecosystem to reproduce the benchmark and validate these performance claims.
What 1M TPS Means for the NEAR Ecosystem and the Blockchain Economy
This milestone validates NEAR as the truly scalable, future-proof protocol for the next era of blockchain applications. With the sharded smart contract implementation and dynamic resharding in development, NEAR protocol is strongly positioned to support increasing cross-chain volume through NEAR Intents, onchain AI agents executing thousands of micro-transactions per second, and enterprise-grade systems leveraging confidential AI compute at global scale.
Importantly, this scaling milestone was achieved on accessible hardware. No supernodes. No proprietary or centralized sequencing layer. No hidden monolithic bottlenecks. With NEAR’s minimal hardware requirements and increasing number of validator seats, we aim to ensure the decentralization and diversity of NEAR’s validator network.
What’s Next for NEAR Scalability
This benchmark is not hypothetical lab code. The majority of the core code used to demonstrate 1M TPS is already on mainnet, with the remainder to be released in version 2.12, enabling validators and the wider NEAR ecosystem to benefit from the many technical optimizations achieved during this performance benchmarking.
NEAR’s scaling work continues. Going forward, we are focused on the following:
- Dynamic resharding on mainnet
- Sharded smart contracts
- Further execution and trie optimizations
- Decoupling consensus and execution for even faster block time
- Scalable RPC infrastructure
- More granular performance reports and additional public benchmarks
The vision of NEAR protocol is simple: a blockchain that can scale indefinitely while remaining decentralized, easy to build on, cost-efficient, and most of all secure. Today’s 1M TPS milestone is a major step toward that future and not the limit—NEAR will continue to scale even further horizontally to support the growing demand of the multi-chain ecosystem and the fast-emerging AI economy.
