Edge-cached reads
01Hot read paths serve from the nearest of 12 regions. Cache invalidation tracks block height, not TTL. Stale reads do not happen.
RPC Infrastructure
Production RPC, sized to your throughput.
37 millisecond median response from 12 regions. WebSocket, archive, and trace endpoints. Dedicated lanes so production traffic never queues behind a shared pool, and edge-cached reads that track block height rather than expiring on a timer.
37ms
Median response
Global p50, 90-day rolling
12
Regions worldwide
Edge-cached reads
99.99%
Endpoint uptime
Production SLA
50+
Networks served
L1 · L2 · DA · AVS
What you get
RPC that holds latency under load.
Hot paths cached at the edge. Dedicated lanes for production. Archive and trace for everything the protocol supports. One uptime SLA across all of it.
Dedicated throughput lanes
02Reserve capacity in requests-per-second tiers. Your traffic runs in its own queue. A noisy neighbor on the shared pool does not affect your latency.
WebSocket and pub/sub
03Subscribe to new blocks, pending transactions, and contract logs. Connection pooling and automatic reconnection handled at the gateway, not your client.
Archive and trace methods
04Full method coverage including eth_call against historical state, debug_traceTransaction, and trace_block. Available on dedicated and cluster tiers.
Smart routing with failover
05Requests route to the lowest-latency healthy node, then fall back across regions on failure. Reorg-aware so cross-region reads stay consistent.
Usage metering and budgets
06Per-method, per-endpoint, per-key metering in real time. Spend caps, alerts, and budget exports available through the dashboard and API.
How it works
Five layers between your request and chain state.
Every request crosses the same path: gateway, balancer, node fleet, observability, and a 24/7 operations desk. The path is the SLA.
- 01
Edge gateways
Auth, rate limiting, caching, and routing run at 12 regions worldwide. TLS terminates close to the client to keep latency under budget.
- 02
Smart load balancer
Health-aware routing distributes requests across nodes and regions. A latency budget enforces per-request response targets.
- 03
Node fleet
Multiple full nodes per chain per region, with archive and trace nodes where the network supports them. Hardware sized to peak traffic, not average.
- 04
Observability
Per-request tracing, p50, p95, and p99 latency, error rates, and per-method usage exposed through the customer dashboard and exported to your stack.
- 05
Operations desk
Engineers respond to incidents in under 15 minutes. Post-mortems on every sev-1 publish within 48 hours.
Production workloads
What teams ship on this stack.
From DEX frontends to MEV searchers to oracle networks, the same RPC fleet powers production workloads with different latency, archive, and budget constraints.
DEX frontends and aggregators
Sub-100ms reads from the user's nearest region. Quote and route generation that does not block on RPC. Geographic routing for regulated jurisdictions.
Trading and MEV systems
Dedicated lanes for low-latency reads. Pending transaction streams through colocated WebSocket endpoints. Predictable performance during volatility.
Wallets and consumer apps
99.99% endpoint uptime. Geographic routing for compliance constraints. Per-key budget controls so a viral moment does not turn into a runaway invoice.
Indexers and data products
Archive and trace access for historical reconstruction. Backfill rates that do not pressure your production reads.
Security and forensics
Full trace coverage with replayable execution. Dedicated capacity so anomaly detection runs without competing for cache space with other tenants.
Oracle and price feeds
Stable read latency under load. WebSocket subscriptions on price-relevant events. No throttling during high-volume publish windows.
RPC tier comparison
Free tier vs shared vs dedicated vs cluster.
Pick the tier that fits the production constraint. Switch tiers as traffic grows, without changing endpoints or refactoring auth.
| Metric | Free tier RPC | Shared pool | Dedicated lane | Cluster |
|---|---|---|---|---|
| Throughput | Best-effort | Burstable, capped | Reserved RPS | Custom hardware |
| Latency profile | Variable | Stable at p50, tails on p99 | Stable across p50 and p99 | Optimized end-to-end |
| Tenancy | Multi-tenant | Multi-tenant | Single-tenant queue | Single-tenant infrastructure |
| Archive + trace | No | Limited | Yes | Yes, custom binaries |
| Uptime SLA | None | 99.9% | 99.99% | Custom |
| Support | Community | Email + docs | Named engineer | Named engineer + on-site review |
| Best for | Hackathons, prototypes | Dev, staging, low-volume prod | Production at scale | Compliance, low-latency reads |
Service specs
What ships with every RPC contract.
Performance and method coverage written into the SLA. Per-network deviations documented explicitly during scoping.
- Per-method, per-region p50 and p99 latency targets
- WebSocket pub/sub with gateway-side reconnection
- Archive and trace methods on dedicated and cluster tiers
- Real-time spend caps, alerts, and budget exports
Specs
- Median latency
- 37ms
- p99 latency
- < 120ms
- Regions
- 12
- Uptime SLA
- 99.99%
- Networks
- 50+
- Methods
- Standard + archive + trace
- Transport
- HTTPS + WebSocket
- Auth
- API key + JWT + IP allowlist
Related infrastructure
What runs behind the RPC fleet.
Frequently asked
RPC questions from production teams.
Don't see your question? A named engineer replies within one business day.
RPC infrastructure refers to the network of nodes and gateways that serve JSON-RPC and WebSocket requests for blockchain data. Applications, wallets, and trading systems read chain state and submit transactions through RPC endpoints. InfraSingularity operates production RPC across 50+ networks with a 37ms global median response time and a 99.99% uptime SLA.
Shared RPC routes traffic through multi-tenant pools where capacity is burstable but variable under load. Dedicated RPC reserves throughput in a single-tenant queue, so request latency stays stable regardless of other customers' traffic. InfraSingularity offers both tiers under one SLA framework.
Yes. Full eth_subscribe support on EVM chains including newHeads, logs, and newPendingTransactions. Native pub/sub on non-EVM networks. Connection pooling and automatic reconnection run at the gateway, so your client does not handle session state.
All standard JSON-RPC methods plus archive and trace methods including debug_traceTransaction, trace_block, eth_call against historical state, and protocol-specific debug methods. Method availability documented per network.
Median, p95, and p99 response time at the gateway across all production traffic, broken down by region and method. Real-time numbers expose on the customer dashboard. The 90-day p50 across all networks currently runs at 37 milliseconds.
Yes. Archive endpoints serve historical state from genesis on all major networks where the protocol supports archive sync. Available on dedicated and cluster tiers. Pricing scales with archive query volume rather than total state size.
Test endpoints provision within minutes through the contact form. Production keys with custom rate limits issue within one business day of contract sign-off. Dedicated lane provisioning takes up to 48 hours from contract finalization.
Put your workload on infrastructure that's built to run it.
One conversation. A named engineer. A reply within one business day. No custody transfer. No lock-in.