Before you approve an OpenClaw rollout for your team, you need a number your finance team can work with. Not “it depends,” not “somewhere between $5 and $500.” A real per-seat figure tied to your team size and usage pattern.
The single biggest cost mistake teams make: planning for individual costs and multiplying by headcount. That math is wrong because it ignores the architecture decision that determines everything else — whether your team shares one OpenClaw instance or runs separate ones per person.
Here is the short answer. A 10-person team on shared infrastructure with model routing spends $180-$400 per month total, or $18-$40 per seat. A 10-person team where everyone runs their own instance spends $600-$1,500 per month. The architecture choice alone creates a 3-4x cost difference.
Where the Money Goes
OpenClaw is open-source and free. You pay for two things: the server it runs on, and the AI models it calls when it works. For a single user, these costs are predictable. For teams, they compound in ways that are not obvious until you see the invoice.
Three cost components matter at team scale:
- Infrastructure — the VPS, local server, or managed platform running your OpenClaw instance(s)
- API tokens — every message, every heartbeat, every skill invocation burns tokens from Anthropic, OpenAI, or Google
- Overhead multipliers — the heartbeat cycle, context injection, and multi-channel routing that generate API calls whether anyone is actively using the agent or not
The overhead multipliers are where teams get surprised. A single OpenClaw instance on Claude Opus 4.6 injects roughly 9,600 tokens of context per turn. At $5 per million input tokens, that is $0.048 per message before anyone types anything. The 30-minute heartbeat runs 48 times per day, and each one is a full API call. On Opus with two channels, heartbeat alone costs $130-$300 per month per instance.
For an individual, that heartbeat cost is manageable. For a team running 10 separate instances, it is $1,300-$3,000 per month in heartbeats alone — doing nothing but checking whether tasks need attention.
Shared Instance vs. Individual Instances
This is the decision that shapes your entire budget. Every other number follows from it.
Shared instance: one OpenClaw deployment serves the whole team. Everyone connects through the same Telegram group, Slack channel, or API endpoint. The agent maintains one memory, one heartbeat cycle, and one set of context. Infrastructure cost: one server. Heartbeat cost: one cycle.
Individual instances: each team member gets their own OpenClaw. Separate memory, separate heartbeat, separate API keys. The agent can be personalized per person but costs scale linearly with headcount.
Hybrid: a shared instance handles team-wide tasks (monitoring, scheduled reports, shared workflows) while 2-3 power users run personal instances for individual work.
| Factor | Shared Instance | Individual Instances |
|---|---|---|
| Infrastructure cost | 1x (one server, scaled up) | Nx (one server per person) |
| Heartbeat cost | 1x | Nx |
| Memory/context | Shared across team | Personal per user |
| Privacy | Low — everyone sees everything | High — fully isolated |
| Customization | Team-level only | Per-person skills and prompts |
| Concurrent usage | Queue contention at scale | No contention |
For most teams under 15 people doing operational work (scheduling, research, email triage, report generation), a shared instance is the right call. The cost savings are substantial and the tradeoffs are manageable. Teams doing sensitive work (legal, HR, executive comms) or teams over 25 where queue contention becomes real should consider individual instances or the hybrid model.
Budget Templates by Team Size
These templates assume self-hosted deployment on a VPS with model routing enabled (Claude Sonnet 4.6 for standard tasks, Opus 4.6 for complex reasoning, GPT-5.4 for high-volume low-stakes tasks). All figures are monthly.
5-Person Team
Recommended architecture: Shared instance
| Component | Cost |
|---|---|
| VPS (2 vCPU, 4 GB RAM) | $12-$20 |
| API tokens (moderate usage, model routing) | $60-$150 |
| Heartbeat (1 instance, Sonnet default) | $25-$40 |
| Monitoring/backups | $5-$10 |
| Total | $102-$220 |
| Per seat | $20-$44 |
At 5 people, individual instances make no financial sense. The infrastructure alone would be $60-$100 before a single API call.
10-Person Team
Recommended architecture: Shared instance with scaled server
| Component | Cost |
|---|---|
| VPS (4 vCPU, 8 GB RAM) | $24-$40 |
| API tokens (moderate-heavy usage, model routing) | $120-$300 |
| Heartbeat (1 instance, Sonnet default) | $25-$40 |
| Monitoring/backups | $10-$15 |
| Total | $179-$395 |
| Per seat | $18-$40 |
Compare to 10 individual instances: $600-$1,500/month. The shared model saves 60-75%.
25-Person Team
Recommended architecture: Hybrid — one shared instance for team workflows, 3-5 individual instances for power users
| Component | Cost |
|---|---|
| Primary VPS (4 vCPU, 16 GB RAM) | $40-$60 |
| Power user VPS (3-5 instances, 2 vCPU each) | $30-$75 |
| API tokens (shared instance, heavy routing) | $200-$450 |
| API tokens (power user instances) | $120-$300 |
| Heartbeat (4-6 instances, mixed models) | $80-$160 |
| Monitoring/backups | $15-$25 |
| Total | $485-$1,070 |
| Per seat | $19-$43 |
At 25 people on a fully shared instance, queue contention becomes noticeable during peak hours. The hybrid model solves this by offloading power users to dedicated instances while keeping the majority on the shared one.
50-Person Team
Recommended architecture: Hybrid with department-level instances
| Component | Cost |
|---|---|
| 3-4 department instances (4 vCPU, 8 GB each) | $72-$160 |
| 5-8 individual power user instances | $50-$120 |
| API tokens (department instances, heavy routing) | $400-$900 |
| API tokens (power user instances) | $200-$500 |
| Heartbeat (8-12 instances, Sonnet default) | $200-$400 |
| Monitoring/backups/management | $30-$50 |
| Total | $952-$2,130 |
| Per seat | $19-$43 |
Per-seat cost stays remarkably flat as teams grow from 5 to 50 because the shared infrastructure absorbs most of the overhead. The main variable is how many power users need dedicated instances.
API Cost Allocation Across Your Team
Once you have a shared instance, someone has to pay the API bill. Three approaches work:
Flat allocation: divide total API cost equally across team members. Simple. Fair if usage is roughly even. Unfair if the marketing team sends 200 messages a day and engineering sends 20.
Usage-based chargeback: track messages per user or per department. Anthropic, OpenAI, and Google all provide usage dashboards, but OpenClaw does not natively tag API calls by user. You will need to either modify the codebase to log per-user usage or estimate based on channel activity.
Department budgets with caps: assign each department a monthly token budget. Set spending alerts through your provider’s console (Anthropic has spend caps, OpenAI has monthly budget limits, Google Cloud has budget alerts). When a department hits 80% of their cap, they get a warning. At 100%, route their requests to a cheaper model instead of cutting them off.
The third option tends to work best for teams over 10 people. It prevents runaway costs without creating friction. A marketing team that burns through their Sonnet budget gets automatically routed to GPT-5.4 for the rest of the month rather than losing access entirely.
Cost Optimization at Team Scale
Individual cost-saving tricks are well documented. At team scale, three strategies have outsized impact:
Model routing with team-level defaults. Set Sonnet 4.6 as the default for all team members. Reserve Opus 4.6 for tasks explicitly tagged as complex reasoning. Route heartbeats and routine checks through the cheapest viable model (Gemini 3.1 Flash or GPT-5.4). On a 10-person team, switching heartbeats from Opus to Sonnet alone saves $80-$200/month.
Heartbeat frequency tuning. The default 30-minute heartbeat is aggressive for most team use cases. If your team works standard business hours, schedule heartbeats every 30 minutes during 8am-6pm and every 2 hours overnight. This cuts heartbeat API calls by 40-50% with no impact on responsiveness during working hours. For teams in a single timezone, this is an immediate win. Multi-timezone teams need to map heartbeat windows to their coverage hours. See our OpenClaw heartbeat scheduling guide for the exact cron configuration.
Prompt caching. Anthropic offers prompt caching that reduces repeated context costs by up to 90%. For a shared instance where the same system prompt and skill descriptions load on every call, caching eliminates the largest per-message overhead. At 9,600 tokens of context per turn across 500+ daily team messages, prompt caching can save $50-$150/month.
Infrastructure Scaling: What Changes at Each Tier
Infrastructure does not scale linearly. A shared instance for 25 people does not need 25x the resources of a single-user setup.
| Team Size | Recommended Specs | Estimated VPS Cost |
|---|---|---|
| 1-5 | 2 vCPU, 4 GB RAM, 40 GB SSD | $12-$20/month |
| 6-15 | 4 vCPU, 8 GB RAM, 80 GB SSD | $24-$40/month |
| 16-30 | 4 vCPU, 16 GB RAM, 160 GB SSD | $40-$60/month |
| 31-50 | 8 vCPU, 32 GB RAM, 320 GB SSD | $60-$100/month |
The RAM matters more than CPU for OpenClaw. Memory holds the context window, conversation history, and skill definitions. CPU spikes during tool execution but idles between requests. If you are choosing between more cores and more RAM, choose RAM.
Storage grows faster than you expect. OpenClaw’s memory system, daily logs, and compacting files accumulate at roughly 50-100 MB per active user per month. Budget 2-4 GB per user per year, and set up log rotation early. Our OpenClaw Docker deployment guide covers volume management for persistent storage.
For teams already running on Hostinger VPS, their KVM 4 plan ($15.99/month, 4 vCPU, 8 GB RAM) comfortably handles 10-15 concurrent users on a shared instance. Beyond that, move to their KVM 8 or consider splitting into department-level instances.
Managed vs. Self-Hosted for Teams
Self-hosting makes sense when your team includes someone who can maintain the deployment: updates, monitoring, troubleshooting Node.js issues, and managing API keys. The OpenClaw self-hosted vs managed comparison covers this in detail for individual users, but the math shifts for teams.
At team scale, the maintenance burden multiplies. Updates need to be coordinated across instances. Downtime affects the whole team. Security patches on the OpenClaw codebase (which Palo Alto Networks has flagged as a risk surface) need to be applied promptly across every deployment.
Self-hosted team cost: infrastructure + API + 4-8 hours/month of DevOps time per instance. At $75/hour for a mid-level DevOps engineer, that is $300-$600/month in labor per instance.
Managed team cost: OpenClaw Cloud at $59/month per instance + API costs. No maintenance labor.
For teams running 3+ instances, managed hosting often costs less than self-hosting once you factor in labor. The breakeven point is roughly 1 hour of DevOps time per instance per month at $60/hour — and most self-hosted instances need more than that.
Frequently Asked Questions
How much does OpenClaw cost per team member?
Between $18 and $44 per seat per month on a shared instance with model routing, regardless of whether your team has 5 or 50 people. Individual instances push that to $60-$150 per seat because infrastructure and heartbeat costs multiply per person.
Should my team share one OpenClaw instance or run separate ones?
Share one instance if your team is under 15 people and does not handle sensitive personal data. The cost savings are 60-75% compared to individual instances. Switch to a hybrid model (shared + a few individual instances for power users) once queue contention becomes noticeable, typically around 20-25 active users on a single deployment.
How do I set spending limits for team OpenClaw usage?
Set caps at the provider level. Anthropic Console lets you configure spend caps per API key. OpenAI has monthly budget limits in their dashboard. Google Cloud has budget alerts with automatic actions. Create separate API keys per department if you need granular tracking, or use a single key with provider-level caps if flat budgeting is sufficient.
What happens to costs when we add the 11th or 26th person?
Per-seat cost stays roughly flat between $18-$44 as you scale. Infrastructure steps up at natural thresholds (you will upgrade your VPS at around 15 users and again around 30), but those upgrades cost $15-$40/month — spread across the new users, the per-seat increase is negligible. API costs grow linearly with actual usage, not headcount.
Can I use cheap models for routine tasks and premium models only when needed?
Yes, and you should. Configure model routing to send heartbeats and simple queries to GPT-5.4 ($1.25 per million input tokens) or Gemini 3.1 Flash while reserving Claude Opus 4.6 ($5 per million input) for tasks that explicitly require deep reasoning. Our OpenClaw multi-model configuration guide walks through the exact settings. On a 10-person team, this typically reduces API costs by 40-60%.
How do heartbeat costs scale with more users?
Heartbeat costs scale per instance, not per user. A shared instance running one heartbeat cycle costs $25-$40/month on Sonnet 4.6. Ten separate instances run ten heartbeat cycles: $250-$400/month. This is the single strongest financial argument for shared instances at team scale.
What infrastructure do I need for a 25-person team?
A shared instance for 25 people needs a VPS with 4 vCPU, 16 GB RAM, and 160 GB SSD — roughly $40-$60/month from providers like Hostinger or Hetzner. If you run a hybrid setup with 3-5 additional power user instances, add $30-$75 for their smaller VPS allocations. Total infrastructure for 25 people: $70-$135/month.
Key Takeaways
- OpenClaw team costs run $18-$44 per seat per month on shared infrastructure with model routing. Individual instances push that to $60-$150 per seat.
- The shared vs. individual instance decision drives a 3-4x cost difference. Default to shared for teams under 15; hybrid for 15-50.
- Heartbeat costs are the silent budget killer at team scale. One shared instance pays for one heartbeat cycle. Ten instances pay for ten.
- Set provider-level spending caps per department. Route departments that hit their budget to cheaper models instead of cutting access.
- Infrastructure scales sublinearly: a 50-person team needs roughly 4-5x the server resources of a 5-person team, not 10x.
SFAI Labs
