How Canton Handles 10,000 TPS Without Sacrificing Privacy

The standard metric for blockchain performance is transactions per second. Ethereum does ~12 TPS on L1. Solana peaks around 2,000 TPS in practice. Visa handles 65,000+ TPS. Canton Network does not compete on this axis — and that is a deliberate architectural choice, not a limitation. Canton's design eliminates the concept of a global TPS ceiling entirely. Instead of every node processing every transaction, Canton distributes work across domains, participants, and synchronizers so that throughput scales linearly with infrastructure. The result: a network processing $8 trillion in monthly RWA volume and 600,000+ daily transactions without any single component acting as a bottleneck.

Why Global TPS Is the Wrong Metric for Canton

On Ethereum, every validator processes every transaction. The network's TPS is bounded by the slowest validator's throughput. This is a fundamental constraint of global-state blockchains: consensus requires agreement on a single shared ledger, so every node must see everything.

Canton has no global state. Each participant node maintains only the ledger state relevant to its users. A repo trade between JPMorgan and Goldman Sachs is validated by those two institutions' nodes — not by all 800+ validators on the network. A tokenized Treasury settlement at DTCC runs through DTCC's infrastructure and its counterparties, without touching Visa's or Fireblocks's nodes. This is not sharding in the Ethereum sense. It is a fundamentally different computation model where privacy and scalability are the same property.

Domain-Based Horizontal Scaling

Canton organizes its network into synchronization domains (sync domains). Each domain provides transaction ordering, conflict detection, and finality for a set of workflows. When a domain approaches capacity, the network does not slow down — it adds another domain. Workflows are distributed across available domains, and each domain operates independently.

Within a single domain, the architecture decomposes into three horizontally scalable components:

• Sequencer: Establishes total order of all messages within the domain using unique timestamps. Multiple sequencers can run behind a load balancer for high availability and increased throughput. A database-backed sequencer design enables multiple sequencer instances per domain.
• Mediator: Acts as transaction commit coordinator. Registers new requests and finalizes them by collecting confirmations from stakeholders. Uses hot-standby replicas for high availability.
• Topology Manager: Manages participant registration, public key rotation, and knows which parties each participant node hosts.

If any of these components becomes a bottleneck, it can be moved to dedicated compute servers independently. New Canton nodes can be added for linear throughput growth. This is not theoretical — Canton 3.4 (released December 10, 2025) demonstrated the architecture's resilience when 600 nodes transitioned to the new version in under 24 hours with minimal downtime and a high first-attempt success rate.

Parallel Execution Through Privacy

Here is the counterintuitive insight: Canton's sub-transaction privacy model is its primary scalability mechanism. Because each party only receives and validates the sub-transactions relevant to them, the network achieves massive parallelism without explicit sharding protocols.

When Bank A executes a repo trade with Bank B, only those two banks' nodes do computational work for that transaction. Simultaneously, Bank C and Bank D can execute a completely independent trade on the same domain, validated only by their nodes. The transactions do not contend for the same resources because they do not share stakeholders. Conflict-free operations execute concurrently, while only operations that share contract state must be serialized.

This proof-of-stakeholder validation model means Canton's effective throughput is proportional to the number of independent workflows running in parallel — not the raw processing speed of any single node. A network with 800+ institutions running thousands of independent bilateral and multilateral workflows simultaneously achieves aggregate throughput that no global-state blockchain can match, regardless of its per-node TPS.

Canton 3.4: The Performance Upgrade

The December 10, 2025 release of Canton 3.4 and Splice 0.5.0 delivered targeted performance improvements:

• Sequencer processing efficiency: Optimizations enable the Global Synchronizer to handle more transactions per second per domain.
• Smart contract validation: Better topology state validation reduces per-transaction processing overhead.
• Validator onboarding: Validators can join the network faster with reduced computing requirements, lowering the barrier to horizontal scaling.
• Memory optimization: Reduced memory footprint per validator node means the same hardware supports higher throughput.

The real-world impact: Canton now sustains 600,000+ daily transactions with over $350 billion in daily asset movement — a 25% increase quarter-over-quarter. The network recorded 3.5 million Canton Coin-related events in a single day, demonstrating that the infrastructure handles burst loads well above average capacity.

Comparing Architectures: Canton vs. Public Chains

Raw TPS comparisons miss the structural differences:

• Ethereum L1 (12 TPS): Every node validates every transaction. Privacy requires separate L2s or ZK circuits. Scales by moving work off-chain.
• Solana (400-2,000 TPS): Parallel execution via Sealevel runtime, but still requires full state replication. Privacy is minimal. Scales vertically with faster hardware.
• Canton (no fixed ceiling): Only stakeholders validate their sub-transactions. Privacy eliminates unnecessary computation. Scales horizontally by adding domains, nodes, and synchronizers.

Canton's daily $350 billion in asset movement dwarfs the combined value secured by Ethereum and Solana DeFi. The throughput comparison that matters is not transactions per second on a benchmark — it is dollars of real economic activity processed per day in production with institutional privacy and regulatory compliance intact.

What This Means for Network Growth

Canton grew from approximately 200 nodes to 600 nodes between July and December 2025 — a 3x expansion in six months. The protocol handled this growth without degradation because adding participants increases the network's parallel processing capacity rather than adding load to a shared bottleneck. Each new institution brings its own validator nodes, its own compute resources, and its own set of bilateral workflows that execute independently of existing traffic.

With the DTCC Treasury tokenization MVP targeting H1 2026 and JPMorgan's Kinexys expanding JPM Coin natively onto Canton, the next phase of growth will test whether the architecture scales from hundreds of institutions to thousands. The domain-based model, combined with Canton 3.4's efficiency improvements, suggests the infrastructure is ready. The bottleneck for Canton's growth is not throughput — it is institutional onboarding speed.