> infrastructure

What we run for ourselves

We run our own fleet of 50+ nodes across 5+ countries, at 99.982% uptime. It's our proving ground: we break in the load, dashboards, runbooks and on-call rotations on infra we pay for ourselves, before any of it reaches you.

fleet by the numbers

This is the current state, and it's refreshed from the registry rather than from marketing claims.

Metric Value Window
Active nodes 50+ current
Countries 5+ current
Fleet uptime 99.982% trailing 90 days
Baseline bandwidth per node 100 Mbps contractual minimum
Base OS Ubuntu LTS 22.04 / 24.04
Days since last Sev-1 19 at render time

fleet geography

5+ countries across Europe, the Americas and APAC. The distribution isn't accidental: we picked it for regulatory diversity, network-route diversity, and time-zone coverage for follow-the-sun on-call.

node types

It's not one type of workload, so each type has its own alert profile, its own runbooks and escalation order.

Type Count Distribution Profile
Validator 38 CA / DE / US / FI / UK signing-critical, slashing-aware alerts, HSM/KMS
RPC 27 DE / US / BG / CZ latency-critical, p95 SLO, caching
Archive 14 DE / CA / US disk-heavy, multi-TB storage, slow reindex
GPU / inference 22 US / DE / AU A100 / H100, vLLM / Triton, OOM watchdog
Storage / DePIN 23 RU / BR / BE / AT Filecoin / Akash, payout reconciliation
Sequencer / prover 8 US / DE deadline-aware, retry queues

why we keep our own fleet

You can read an SRE book and draw a dashboard. You can sign an SLA on paper. But the only way to know a tool works at 3 AM on day 90 is to run it at 3 AM on day 90. On your own dime.

Every runbook you'd get from us has been through our own fleet first. Every alert is tuned not to wake an engineer for nothing. Every dashboard shows what people actually look at during an incident, not what looks clean in a screenshot.

That's operator discipline, and our own fleet is where we sharpen it before we bring it to clients — because that's the only place we can.

the same engineers run client infra

Same team, same runbooks, same on-call. Just under your workload.

[ See OPERATE ]