[ FIXED PLAN ] →
Migration plan in 48-72h. Target architecture, milestones, rollback strategy, budget, risks.
For when you need to scope the migration before committing to the full program.
> services / scale
Infrastructure engineering and migrations for Web3, AI and ZK
Greenfield builds, cloud → bare-metal migrations, regional splits, hard-fork cutovers, burst capacity. We move running systems. Reversibly. Without losing state or signatures.
> what's included
Scale covers any project that reshapes infrastructure: growth, moves, new regions, hard-forks. The scope lands in a migration plan.
| Scope | Owned by us | Owned by you |
|---|---|---|
| Migration plan: target architecture, dry-runs, rollback | ✓ Owned | - |
| Sourcing new regions and bare-metal partners | ✓ Owned | - |
| Multi-region Kubernetes topology, mesh, DNS cutover | ✓ Owned | - |
| State transfer (chain data, model weights, prover state) | ✓ Owned | - |
| Key ceremony for validator key migration | ✓ Coordinated (we direct) | ✓ Co-owned (keys stay with you) |
| Burst scenarios: 100 nodes in 72h | ✓ Owned | - |
| Business decision to migrate / trade-offs | - (we analyse, recommend) | ✓ Owned |
| Billing for old and new infra during transition window | - | ✓ Owned |
> migration scenarios
Every migration is planned as a reversible process. First a dry-run in an isolated copy of the target environment, then a canary rollout of a subset, then cutover with a rollback that fires within an hour. No "Friday-evening prod jumps".
Cloud → bare-metal
Cost optimisation
Moving GPU workloads or node fleets from public cloud to bare-metal providers. Typical savings: 40-60% on cost-per-token / cost-per-block.
Multi-region split
Latency & resilience
Splitting single-region infra across 3+ regions. Survives regional outages, drops p95 latency for global traffic.
Hard-fork cutover
Chain upgrades
Coordinated validator upgrades for hard-fork blocks. Pre-flight dry-runs, version pinning, rollback playbook in case the majority doesn't fork.
Burst engagement
100 nodes in 72h
Incentivized testnets, fresh network launches, GPU farms for incentive seasons. Sourcing, deployment, monitoring: all in parallel. Aim: top-tier rewards.
Each scenario gets its own runbook, rollback plan and checkpoints. All migrations include a key ceremony for signing systems: HSM/KMS workflow, multi-party approval, audit trail in Git.
> stack we ship with
Web3:
Cosmos SDK
Geth
Reth
OP Stack
Arbitrum Orbit
Polygon CDK
EigenDA
Celestia
AI / LLM:
vLLM
Triton
TensorRT-LLM
NVIDIA H100 / A100
Ray
Kubeflow
ZK:
SP1
RISC Zero
Boundless
Brevis
Jolt
Halo2
DePIN:
Filecoin
Akash
Render
io.net
Gensyn
Platform:
Kubernetes
Terraform
Ansible
Prometheus
Grafana
Loki
OpenTelemetry
PagerDuty
> engagement models
Scale projects are usually time-boxed engagements with fixed milestones.
[ RETAINER ] →
Migration program. Full team engagement for 4-12 weeks, milestone-based payment.
The standard model for cloud → bare-metal or multi-region splits.
[ T&M ] →
Burst engagement. For one-off events: hard-fork, 100-node burst, emergency rebalance.
When urgency is known but scope can't be pinned down in advance.
> security baseline for migrations
Migration is the moment of highest risk for key material. What we always enforce on a scale project.
| Control | What we ensure |
|---|---|
| Key ceremony | Multi-party-approved procedure for all validator keys. Audit trail. Keys never leave your HSM/KMS. |
| HSM / KMS workflow | YubiHSM, AWS KMS, GCP KMS, Vault supported. Signing by process, not by handing over material. |
| Dry-runs | Every migration runs first in an isolated copy of the target environment. |
| Rollback playbook | Pre-built rollback plan with <60 min RTO at every cutover step. |
| Distributed locks | Guarantees single-instance signing per chain throughout the transition window. |
> what we'd build for you
Real migrations and scale projects across the four ICPs.
Testnet → mainnet migration with key ceremony and zero-loss state transfer. Coordinated cutover, signing monitoring before and after, rollback playbook in case of chain split.
Spot-fleet to dedicated H100 cluster, cost-per-token cut by 60%. Model-weight migration, cache warm-up, blue-green cutover for production traffic.
Prover capacity ramped 5x for a proving-incentive season. GPU sourcing in <2 weeks, scheduler expansion, proof-throughput monitoring across the network.
Node fleet rebalanced across 12 regions in 72h to hit reward thresholds. Payout-data analysis, regional sourcing, phased rollout, results reconciliation after one week.
> SLA tiers
After the scale phase, infra moves into operate. Three coverage levels.
| Tier | Response p95 (Sev-1) | Coverage | Incident report | Engineer hours / mo |
|---|---|---|---|---|
| Bronze | 30 min | Business hours, 5×8 | Within 48h | 40 |
| Silver | 15 min | 24/7 on-call rotation | Within 24h | 80 |
| Gold | 5 min | 24/7 with dedicated engineer | Within 12h | 160+ |
> related services
> FAQ
Yes, with a proper key ceremony and distributed locks. The cutover window is usually one missed epoch, with no double-sign and no slashed stake. Specifics depend on the chain.
Cloud → bare-metal for a single application: 4-6 weeks. Multi-region split with DNS cutover: 3-4 weeks. 100-node burst for a testnet: 72h. Hard-fork cutover: a planned window, usually 2-4 weeks of prep.
Every step has a rollback playbook with <60 min RTO. The old stack stays alive until the new one is fully validated. That doubles billing during the transition window, but that's the explicit price of reversibility.
We direct it: we write the procedure, run dry-runs, coordinate participants. Keys stay with you (HSM/KMS); signing happens through your processes. Audit trail in Git.
Yes. Often the brief is "we already have infra, move it for us". If you want the operate phase after migration, we roll into a retainer. If not, we hand off with runbooks and your team takes it from there.