Infrastructure Economics: How the Best Companies Print Money

Infrastructure isn't visible until it fails. Yet it defines the reliability of everything above it.

The companies that build this layer (the APIs that move money, the platforms that route data, the tools that keep systems alive) operate in silence. No consumer brand. No viral growth. No TechCrunch headlines.

But they print money.

I've spent years studying infrastructure businesses. The ones that work share a pattern: high gross margins (70-85%), negative churn (expansion revenue > lost revenue), and compounding moats built on switching costs and technical depth.

Here are five worth studying.

Stripe: The $50B Payment Infrastructure Giant

Stripe processed over $1 trillion in payment volume in 2023. Revenue: ~$16B. Valuation: $50B.

They started with seven lines of code to accept a payment. Then they built everything adjacent: billing, fraud detection, tax compliance, banking APIs, revenue recognition. Once a company runs on Stripe, they're running their entire financial stack on Stripe. Switching means rebuilding from scratch.

The moat isn't the payment processing. It's the integration depth. Every product they add makes leaving harder.

Datadog: 130% Net Revenue Retention

Datadog did $2.1B in revenue in 2023 at 81% gross margins. Their secret: pricing that scales automatically with customer growth.

You start with infrastructure monitoring at $15/host/month. Then you add APM. Then logs. Then security. Then real user monitoring. Average customer spending goes from $75K/year to $500K+/year over 3-4 years. No sales motion required.

Observability tools embed into every service. Ripping out Datadog means rewriting dashboards, alerts, runbooks, and on-call workflows. Most companies never switch.

Plaid: The $13.4B API Layer

Plaid connects 200M+ bank accounts across 8,000+ apps. Revenue: ~$600M. Gross margins: ~80%.

Before Plaid, every fintech had to build direct integrations with thousands of banks. Plaid did it once, sold it to everyone. Network effects compound: more apps means more user data means better reliability means more apps.

Visa tried to acquire them for $5.3B in 2020. DOJ blocked it. Current valuation: $13.4B.

MongoDB: 30%+ Revenue Growth for 7 Straight Years

MongoDB did $1.68B in revenue in FY2024 at 74% gross margins. Their managed cloud service (Atlas) has 10x better unit economics than self-hosted.

Customers start with free tier, grow to $100/month, then $10K/month, then $500K+/year as data scales. The database is the hardest infrastructure layer to migrate. Once production data lives in MongoDB, switching requires schema redesign, application rewrites, and zero-downtime data migration.

Most companies never switch.

Cloudflare: 20% of the Internet

Cloudflare did $1.3B in revenue in 2023 at 77% gross margins. They handle 20%+ of all internet requests.

Their network is a moat. The more traffic they handle, the better their threat intelligence. The better their intelligence, the better their security. The better their security, the more customers they attract. Pricing scales with usage: free tier to $20/month to $5K+/month as traffic grows.

Moving DNS, CDN, and security to a new provider means downtime risk, rewriting security rules, and losing threat intelligence history. Most companies never switch.

The Pattern

High gross margins (70-85%). Usage-based pricing that scales automatically. Negative churn. Compounding moats built on switching costs. Mission-critical infrastructure where downtime equals revenue loss.

The defensibility isn't in the technology. It's in the operational knowledge, the integration depth, and the years of edge cases solved. Competitors can't replicate that quickly.

What This Means for Founders

If you're building infrastructure, ask yourself:

Does my product become more valuable as customers scale? Do switching costs increase over time? Am I solving a mission-critical problem? Can I achieve 70%+ gross margins at scale? Does my pricing model capture customer growth?

If yes, you're building a cash machine.

If no, you're building a feature.