The Great SaaS Exodus: Why Companies Are Self-Hosting Again

Basecamp spent $3.2 million on cloud infrastructure in 2022. In October 2022, they announced they're leaving the cloud entirely, expecting to save $7 million over five years.

37signals (Basecamp's parent company) isn't alone. Companies across industries are reversing a decade of "cloud-first" doctrine and bringing infrastructure back in-house.

This isn't nostalgia for 2010s-era server racks. It's a calculated response to SaaS pricing models that scale faster than revenue.

The Numbers Behind the Exodus

Public Migrations (Announced 2022-2026)

37signals (Basecamp, HEY):

• Previous AWS spend: $3.2M/year
• New infrastructure: $840K (hardware) + $286K/year (operations)
• 5-year savings: $7M
• Migration timeline: 18 months
• Announcement: October 2022, blog post

Discord:

• Migrated message database from MongoDB Atlas to self-hosted ScyllaDB
• Previous cost: $150K/month ($1.8M/year)
• New cost: $50K/month ($600K/year)
• Savings: $1.2M/year
• Performance gain: 90% reduction in latency

Dropbox:

• Migrated from AWS S3 to custom infrastructure (2016, but trend continued)
• Saved $75M over 2 years
• Reduced per-user storage cost from $5 to $2

Coinbase:

• Self-hosts core exchange infrastructure
• Uses AWS only for ancillary services
• Estimated savings: $100M+ over 5 years
• Reason: Compliance, data sovereignty, cost

Private Migrations (Industry Reports)

Gartner survey (2025): 47% of enterprises are "repatriating" cloud workloads back to on-premise or self-hosted infrastructure.

Reasons cited:

1. Cost overruns (68%)
2. Data sovereignty requirements (42%)
3. Performance issues (31%)
4. Vendor lock-in concerns (29%)

A16Z report (2024): Cloud costs now represent 30-50% of revenue for SaaS startups, up from 15-20% in 2019.

Why SaaS Pricing Broke

The Promise (2010-2015)

SaaS vendor pitch:

• "Pay only for what you use"
• "No upfront infrastructure costs"
• "Scales automatically with your business"
• "Cheaper than hiring DevOps team"

This was true... initially.

The Reality (2020-2026)

What actually happened:

1. Usage-based pricing became a tax on growth

Example: Vercel pricing for Next.js hosting

• Free tier: 100GB bandwidth
• Pro: $20/month + $40 per 100GB overage
• Company with 2TB bandwidth/month: $20 + ($40 × 19.9) = $816/month

Self-hosted equivalent:

• Hetzner dedicated server: $45/month
• 20TB included bandwidth
• Savings: $771/month ($9,252/year)

2. Per-user pricing doesn't match value delivered

HubSpot Sales Hub: $100/user/month

• 10-person sales team: $12,000/year
• 50-person sales team: $60,000/year

Value delivered doesn't scale linearly with users, but pricing does.

Self-hosted EspoCRM:

• Cost: $850/year (regardless of users)
• 10 users: Saves $11,150/year
• 50 users: Saves $59,150/year

3. "Free tier" lures, then pricing jumps

Slack pricing evolution:

• 2014: Free tier with full search history
• 2018: Free tier limited to 10,000 messages (90 days for most teams)
• 2022: Standard plan $6.67/user/month
• 2024: Pro plan required for features once included in Standard
• 10-person team cost: $800/year → $2,400/year (200% increase)

4. Enterprise features held hostage

Common SaaS pattern:

• Starter: $10/user (missing SAML SSO, audit logs)
• Professional: $25/user (adds SSO, still no audit logs)
• Enterprise: "Contact sales" ($100+/user)

Security and compliance become upsell opportunities.

Case Study: Mid-Size SaaS Company (150 Employees)

SaaS Stack (Annual Costs)

| Service | Tool | Cost/Year | | ------------------ | ----------------------- | ----------------- | | Communication | Slack Business+ | $20,250 | | Code hosting | GitHub Teams | $25,200 | | CI/CD | CircleCI | $12,000 | | Monitoring | Datadog | $36,000 | | CRM | HubSpot | $54,000 | | Project management | Asana Business | $29,970 | | Cloud hosting | AWS (database, compute) | $180,000 | | Error tracking | Sentry Business | $8,400 | | Analytics | Mixpanel Growth | $18,000 | | Total | | $383,820/year |

Self-Hosted Equivalent

| Service | Tool | Cost/Year | | ------------------ | --------------------------- | ----------------------- | | Communication | Mattermost self-hosted | $1,440 (infrastructure) | | Code hosting | GitLab CE | $1,800 | | CI/CD | GitLab CI (included) | $0 | | Monitoring | Prometheus + Grafana | $720 | | CRM | EspoCRM | $1,200 | | Project management | Taiga | $720 | | Hosting | Dedicated servers (Hetzner) | $28,800 | | Error tracking | Sentry self-hosted | $720 | | Analytics | Plausible self-hosted | $600 | | DevOps labor | 1 FTE @ $120K/year | $120,000 | | Total | | $156,000/year |

Savings: $227,820/year (59% reduction)

ROI: Even after hiring a dedicated DevOps engineer, saves nearly $230K annually.

What Changed: Docker Revolution

Pre-Docker Era (2005-2015)

Self-hosting required:

• Deep Linux knowledge
• Manual dependency management
• Custom deployment scripts
• Configuration drift across environments
• "Works on my machine" problems

Deployment complexity = Barrier to entry

Result: SaaS won because it was genuinely easier.

Post-Docker Era (2015-Present)

Self-hosting now:

git clone https://github.com/mattermost/docker
cd docker
docker-compose up -d
# Entire application stack running in 3 commands

What Docker enabled:

• Reproducible deployments
• Dependency isolation
• One-command scaling
• Platform independence

Deployment complexity = Solved problem

Result: SaaS convenience advantage disappeared.

The Sovereignty Awakening

Data Privacy Incidents That Changed Perception

Zoom (April 2020):

• Caught routing traffic through China despite claiming end-to-end encryption
• "Zoombombing" incidents exposed weak security defaults
• Outcome: Enterprises realized they had zero control over vendor practices

SolarWinds (December 2020):

• Supply chain attack compromised 18,000 customers
• Attackers had access for 9 months
• Lesson: SaaS vendor breach = you're breached

Okta (March 2022):

• Lapsus$ hacking group accessed internal systems
• 366 customer companies potentially compromised
• Realization: Identity provider breach = skeleton key to your entire company

LastPass (August 2022):

• Encrypted vaults stolen
• Attackers had 9 months of access to development systems
• Trust destroyed: Password manager, the ultimate trust service, fundamentally compromised

GDPR Awakening (2018-Present)

Before GDPR: "Our vendors handle compliance"

After GDPR:

• €50M fine to Google (2019) for consent violations
• €746M fine to Amazon (2021) for data processing violations
• Realization: Using non-compliant vendor = you're liable

Self-hosting solution:

• Data never leaves EU infrastructure
• No third-party processor = simpler compliance
• Complete audit control

The Technical Maturity Curve

Phase 1: SaaS Adoption (2010-2018)

Characteristics:

• Small team (5-20 employees)
• Focus on product-market fit
• No DevOps expertise
• SaaS makes sense (speed over cost)

Phase 2: SaaS Scaling Pain (2018-2022)

Characteristics:

• Growing team (20-100 employees)
• SaaS costs scaling faster than revenue
• Complexity from 20+ SaaS tools
• Integration hell (Zapier holding everything together)

Phase 3: Selective Self-Hosting (2022-Present)

Characteristics:

• Mature team (50-200+ employees)
• Dedicated DevOps/Platform team
• Strategic tool evaluation
• Hybrid approach: Self-host commodity tools, keep SaaS for specialized services

Migration pattern:

1. First to self-host: GitLab (easy, high ROI)
2. Second: Monitoring (Prometheus/Grafana)
3. Third: Internal communication (Mattermost)
4. Fourth: Project management (Taiga, Plane)
5. Keep SaaS: Email, accounting, payroll

Economic Factors Accelerating Exodus

1. Venture Capital Pullback (2022-Present)

Cheap VC money dried up.

Startups can no longer afford to:

• Burn $500K/year on AWS
• Pay $100/user/month for every SaaS tool
• Ignore unit economics

CFO conversation shift:

• 2020: "Just use SaaS, we'll optimize later"
• 2024: "Why are we paying $400K/year for tools we could run for $50K?"

2. SaaS Vendor Consolidation

Private equity playbook:

1. Buy popular SaaS company
2. Fire half the engineering team
3. Increase prices 30-50%
4. Cut support quality
5. Extract maximum revenue

Recent examples:

• Salesforce acquiring Slack (price increases followed)
• Thoma Bravo buying Mailchimp, Plannuh, Stamps.com (price increases)
• Vista Equity buying Marketo, Pluralsight (quality decline reported)

Customer response: "If quality is declining while prices increase, why not self-host?"

3. AI Commoditization

SaaS value prop used to be:

• "We have data scientists building ML models you can't replicate"

Now:

• Open-source LLMs (Llama, Mistral) match proprietary models
• Self-hosting AI is feasible
• "Smart features" no longer justify 10x pricing premium

Predictions: Where This Goes

Short-term (2026-2028)

SaaS vendors will:

• Introduce "self-hosted" tiers (ironic reversal)
• Lower prices for loyal customers (retention focus)
• Add "data residency" options (EU, US regions)
• Acquire open-source competitors

Companies will:

• Migrate 30-50% of SaaS tools to self-hosted
• Keep SaaS for compliance-heavy services (payments, email)
• Invest in DevOps/platform engineering teams

Long-term (2028-2032)

Hybrid infrastructure becomes standard:

• Self-hosted: Databases, internal tools, analytics
• SaaS: Specialized compliance (payments, email), AI services
• Edge: Static sites, CDN (Cloudflare, Vercel)

New business models emerge:

• Open-source companies offer "managed self-hosting"
• Example: You run GitLab on your infrastructure, GitLab provides support/updates
• Best of both worlds: Data sovereignty + vendor expertise

Developer tooling improves:

• Kubernetes becomes as easy as Docker Compose
• "Deploy button" for self-hosted apps becomes standard
• Infrastructure-as-code templates proliferate

The Exit-Saas Philosophy

This isn't anti-SaaS extremism.

SaaS solved real problems:

• Eliminated upfront infrastructure investment
• Democratized access to enterprise software
• Enabled rapid experimentation

But the pendulum swung too far:

• "Cloud-first" became "cloud-only" without evaluation
• Convenience premium became exploitation
• Data ownership was surrendered without consideration

The correction isn't "self-host everything."

It's selective self-hosting based on:

1. Cost analysis (5-year TCO)
2. Technical capacity (do you have DevOps skills?)
3. Data sensitivity (compliance requirements)
4. Vendor lock-in risk (can you export data easily?)

Companies that thrive:

• Self-host commodity infrastructure
• Use SaaS for specialized services
• Maintain optionality (can switch either direction)

Browse our tools directory for 800+ self-hosted alternatives to popular SaaS platforms.

The future isn't SaaS or self-hosted. It's knowing the difference.