To self-host or not to self-host?

Startup credits are not really what you think, and this might not sit well with a lot of people. But let me explain how self-hosting can be a better way, and how to optimize your costs from first principles while doing so.

As a technical founder/co-founder, the first few months of your startup is all about talking to customers, finding potential pain points to solve, and then creating an MVE for them (Minimum Viable Experience - the age of MVPs is dead, read more here) - in that order. Most platforms are more than happy to give a significant amount of startup credits (infact we do too!)

We don’t want to spread ourselves too thin, XYZ platform is such a godsend!

^^ That’s the most common testimonial (or undisclosed ad?) we see on TL, innit?

But that honeymoon phase ends soon. Then the reality hits: most platforms give these startup credits as a honey-trap. The loop goes like this:

You get a lot of credits -> You build a lot on these platforms -> You get locked in -> Credits expire -> You:

That was me when ALL (I kid you not, ALL) our startup credits expired in a span of one unkind week.

Meanwhile on the other end: You hear your Co-founder/CEO/CFO shouting:

“EXCUSE ME?!”

That’s the expression my co-founder @Uttaranxnayak hits me with when he sees some unholy number of $$ going somewhere.

And that’s where it hits you:

“I should’ve seen the pricing page, dammit!”

To self-host or not to self-host?

That’s why I’m writing this article - maybe you’ll find it relatable as well.

If your tech stack is something more than a Next.js app, most of y’all already will be relying on (but hopefully not fully locked-in to) a cloud provider (think AWS, GCP, Azure, etc.) once you start serving customers and moving to real revenue. And once the startup credits run out, the endless tug-of-war between increasing capabilities and cost-cutting starts.

But the problem is (almost) never cloud as a whole. For an early-stage founder, it’s most likely the last thing you want to switch over. Unless you’re working with EXTREMELY sensitive data and require tons of compliances at every step of the way, keeping a cloud provider is probably a good option.

It’s all the tools your team uses that’s NOT directly in your cloud. Invisible cost sinks that can paralyze the daily operations if not noticed beforehand.

The Documentation workspace that charged you by seats?
The task management platform that ran out of free users and suddenly adds 20% to your monthly technical spend budgets?

Yes, those are the ones.

An Example - Our Own Budget Costs

Back in December 2024, just fresh off our Pre-seed round, we applied for a lot of startup credits - cloud, documentation, task management software - you name it.

A few weeks into December - 15 iterations done, revenue starting to flow, things starting to look good. We got our first positive MIS report. Next week:

“Anuran, WTF - we just got billed $5k this month for tech?!”

“Hey, I am locked out of the company documentation - what’s happening?”
> Checks subscription - expired.
> Sees how to reactivate it? “Pay $2k per month”
> FUUUUUUU…

The total bill = ~$8k/month.

(Me calculating the monthly spend that week.)

That was the worst week in my second-half of 2025 - BY FAR. Startup ops paused without it. We had to rethink. What now?

No more oversight.

It was not a problem anymore - it was an emergency. We reluctantly paid - to get one month of access. Over the next 3 days:

Backed up all our data in GCS (thanks Google Cloud) - took us 2 days.
Started setting up self-hosted equivalents of most (now almost all) subscriptions that passed the “ARISE” criteria.

Took us more than a month - but by January 15th, we’re almost done.

What’s the ARISE Criteria?

I’m a big fan of acronyms, but not the corpo lingo that their full forms entail. So I made up one myself: “ARISE” - Alright, Run Infra (or) Subscribe (to) Everything?

The rule of thumb is pretty comprehensive but intuitive:

Find out the costs of all components, and normalize them on a scale of 0–100. (Highest cost = 100)
Assign an impact score based on how much of the team’s total weekly (if not daily) bandwidth goes into that particular tool and the RoI derived out of it.
Calculate RoI by seeing how much of it goes into actually doing business vs. internal workflows.
Factor in maintenance bandwidth costs for self-hosting.
Consider feature gaps and their impact.
Multiply the scores, sort in descending order, and self-host the ones above the median if self-hosting cost is lower.

Example Scenario (Generated with Claude)

Well, so I asked Claude to generate an example out of it, and it captured the idea perfectly.

(Well, I’m not lazy, I’m just in power-saving mode.)

Here’s the sample scenario:

Team Size: 25 people
Total Monthly Software Cost: $10,000
Work Week: 50 hours

Self-Hosting Base Infrastructure: $400/month DevOps Resource: 1 person (from existing team)

Complete ARISE Analysis

1. Communication Suite (Slack Premium → Mattermost)

Current State

Monthly Cost: $2,000
Normalized Cost Score: 100
Team Usage: 25/25 (100%)
Daily Bandwidth: 4 hours
Bandwidth Impact: 4.0

Self-Hosting Assessment

Alternative: Mattermost
Hosting Cost: $80/month
Maintenance: 6 hours/week
Maintenance Bandwidth: 0.048
Normalized Hosting Cost: 4.0

Feature Gap

Missing: Some integrations, threading UX polish, enterprise search
Coverage: 75% of features available
Workaround Time: 1 hour/day
Feature Gap Impact: 0.25

ARISE Delta: 299.808 ✅

Calculation

(100 * 4.0 * 1.0) - (0.048 * 4.0) - (400 * 0.25)
= 400 - 0.192 - 100
= 299.808

Monthly Savings: $1,920

2. Project Management & Docs (Linear + Notion → Plane + Outline)

Current State

Monthly Cost: $1,500
Normalized Cost Score: 75
Team Usage: 25/25 (100%)
Daily Bandwidth: 3 hours
Bandwidth Impact: 3.0

Self-Hosting Assessment

Alternative: Plane + Outline
Hosting Cost: $50/month
Maintenance: 4 hours/week
Maintenance Bandwidth: 0.032
Normalized Hosting Cost: 2.5

Feature Gap

Missing: Real-time collaboration polish, AI features, some integrations
Coverage: 85% of features available
Workaround Time: 0.5 hours/day
Feature Gap Impact: 0.075

ARISE Delta: 208.045 ✅

Calculation

(75 * 3.0 * 1.0) - (0.032 * 2.5) - (225 * 0.075)
= 225 - 0.08 - 16.875
= 208.045

Monthly Savings: $1,450

3. Customer Support (Zendesk → Chatwoot)

Current State

Monthly Cost: $1,200
Normalized Cost Score: 60
Team Usage: 8/25 (32%)
Daily Bandwidth: 5 hours
Bandwidth Impact: 5.0

Self-Hosting Assessment

Alternative: Chatwoot
Hosting Cost: $40/month
Maintenance: 3 hours/week
Maintenance Bandwidth: 0.024
Normalized Hosting Cost: 2.0

Feature Gap

Missing: Advanced automation, some integrations, advanced reporting
Coverage: 80% of features available
Workaround Time: 0.75 hours/day
Feature Gap Impact: 0.15

ARISE Delta: 81.552 ✅

Calculation

(60 * 5.0 * 0.32) - (0.024 * 2.0) - (96 * 0.15)
= 96 - 0.048 - 14.4
= 81.552

Monthly Savings: $1,160

4. Email & Calendar (Google Workspace → Mailcow + Nextcloud)

Current State

Monthly Cost: $500
Normalized Cost Score: 25
Team Usage: 25/25 (100%)
Daily Bandwidth: 2 hours
Bandwidth Impact: 2.0

Self-Hosting Assessment

Alternative: Mailcow + Nextcloud
Hosting Cost: $70/month
Maintenance: 7 hours/week (email is critical)
Maintenance Bandwidth: 0.056
Normalized Hosting Cost: 3.5

Feature Gap

Missing: Spam filtering quality, mobile app polish, reliability SLA
Coverage: 75% of features available
Workaround Time: 0.8 hours/day
Feature Gap Impact: 0.2

ARISE Delta: 39.804 ✅

Calculation

(25 * 2.0 * 1.0) - (0.056 * 3.5) - (50 * 0.2)
= 50 - 0.196 - 10
= 39.804

Monthly Savings: $430

5. CRM & Sales (HubSpot Starter → SuiteCRM)

Current State

Monthly Cost: $600
Normalized Cost Score: 30
Team Usage: 7/25 (28%)
Daily Bandwidth: 3 hours
Bandwidth Impact: 3.0

Self-Hosting Assessment

Alternative: SuiteCRM
Hosting Cost: $35/month
Maintenance: 4 hours/week
Maintenance Bandwidth: 0.032
Normalized Hosting Cost: 1.75

Feature Gap

Missing: Email sequences, marketing automation, modern UX
Coverage: 70% of features available
Workaround Time: 1 hour/day
Feature Gap Impact: 0.3

ARISE Delta: 17.584 🟡

Calculation

(30 * 3.0 * 0.28) - (0.032 * 1.75) - (25.2 * 0.3)
= 25.2 - 0.056 - 7.56
= 17.584

Status: Borderline - Evaluate based on sales team tolerance for UX gaps

6. Analytics & Monitoring (Mixpanel + Datadog → Plausible + Prometheus + Grafana)

Current State

Monthly Cost: $800
Normalized Cost Score: 40
Team Usage: 10/25 (40%)
Daily Bandwidth: 1 hour
Bandwidth Impact: 1.0

Self-Hosting Assessment

Alternative: Plausible + Prometheus + Grafana
Hosting Cost: $60/month
Maintenance: 5 hours/week
Maintenance Bandwidth: 0.04
Normalized Hosting Cost: 3.0

Feature Gap

Missing: Some advanced funnels, alert sophistication
Coverage: 85% of features available
Workaround Time: 0.3 hours/day
Feature Gap Impact: 0.045

ARISE Delta: 15.16 🟡

Calculation

(40 * 1.0 * 0.4) - (0.04 * 3.0) - (16 * 0.045)
= 16 - 0.12 - 0.72
= 15.16

Status: Borderline - Depends on data volume growth trajectory

7. Password Manager (1Password → Vaultwarden)

Current State

Monthly Cost: $300
Normalized Cost Score: 15
Team Usage: 25/25 (100%)
Daily Bandwidth: 0.25 hours
Bandwidth Impact: 0.25

Self-Hosting Assessment

Alternative: Vaultwarden (Bitwarden)
Hosting Cost: $20/month
Maintenance: 2 hours/week
Maintenance Bandwidth: 0.016
Normalized Hosting Cost: 1.0

Feature Gap

Missing: Travel mode, advanced admin controls
Coverage: 95% of features available
Workaround Time: 0.05 hours/day
Feature Gap Impact: 0.0025

ARISE Delta: 3.725 ❌

Calculation

(15 * 0.25 * 1.0) - (0.016 * 1.0) - (3.75 * 0.0025)
= 3.75 - 0.016 - 0.009375
= 3.725

Recommendation: Keep subscription - Security-critical, minimal savings

8. Code Repository (GitHub → GitLab CE)

Current State

Monthly Cost: $400
Normalized Cost Score: 20
Team Usage: 8/25 (32%)
Daily Bandwidth: 0.5 hours
Bandwidth Impact: 0.5

Self-Hosting Assessment

Alternative: GitLab CE + Runners
Hosting Cost: $90/month (CI/CD compute-heavy)
Maintenance: 6 hours/week
Maintenance Bandwidth: 0.048
Normalized Hosting Cost: 4.5

Feature Gap

Missing: GitHub Actions marketplace, Dependabot, Copilot integration
Coverage: 90% of features available
Workaround Time: 0.2 hours/day
Feature Gap Impact: 0.02

ARISE Delta: 2.92 ❌

Calculation

(20 * 0.5 * 0.32) - (0.048 * 4.5) - (3.2 * 0.02)
= 3.2 - 0.216 - 0.064
= 2.92

Recommendation: Keep subscription - Developer experience premium worth cost

9. Cloud Infrastructure (AWS → Bare Metal)

Current State

Monthly Cost: $1,800
Normalized Cost Score: 90
Team Usage: 5/25 (20%)
Daily Bandwidth: 1.5 hours
Bandwidth Impact: 1.5

Self-Hosting Assessment

Alternative: Dedicated servers
Hosting Cost: $1,200/month
Maintenance: 8 hours/week
Maintenance Bandwidth: 0.064
Normalized Hosting Cost: 60

Feature Gap

Missing: Managed services, auto-scaling, global CDN
Coverage: 60% of managed features needed
Workaround Time: 2 hours/day
Feature Gap Impact: 0.8

ARISE Delta: 1.56 ❌

Calculation

(90 * 1.5 * 0.2) - (0.064 * 60) - (27 * 0.8)
= 27 - 3.84 - 21.6
= 1.56

Recommendation: Keep subscription - Managed services essential for product velocity

10. Design Tools (Figma → Penpot)

Current State

Monthly Cost: $900
Normalized Cost Score: 45
Team Usage: 6/25 (24%)
Daily Bandwidth: 6 hours
Bandwidth Impact: 6.0

Self-Hosting Assessment

Alternative: Penpot
Hosting Cost: $30/month
Maintenance: 2 hours/week
Maintenance Bandwidth: 0.016
Normalized Hosting Cost: 1.5

Feature Gap

Missing: Plugins, advanced prototyping, FigJam, dev handoff quality
Coverage: 50% of features available (significant gaps)
Workaround Time: 2.5 hours/day
Feature Gap Impact: 1.25

ARISE Delta: -16.224 ❌

Calculation

(45 * 6.0 * 0.24) - (0.016 * 1.5) - (64.8 * 1.25)
= 64.8 - 0.024 - 81
= -16.224

Recommendation: Keep subscription - Feature gaps would reduce designer productivity by 40%

Median ARISE Delta: 16.372

[In hindsight, that’s common sense, but you won’t believe how few people actually do this!]

How our current setup looks like

We followed this procedure, and moved to self-hosting stuff on our GCP infra - $6k of monthly costs became ~$3k, now well on its way to $2k.

What we self-host: Newsletter Manager, N8N, project management system, internal documentation management, our main product, etc.

What we don’t self-host: Comms, Emailing infrastructure, designing software, LLM APIs.

Conclusion

Now that you’ve reached the end of the article - congratulations! You’ve successfully read through the most boring article I’ve ever written! Give yourself a pat in the back, and maybe comment if this helped you?