Disappearing Infrastructure

Something interesting is happening to infrastructure.

It’s not going away. It’s becoming invisible.

Complexity is being pushed down into runtimes and platforms. Things that used to require setup, libraries, and careful configuration now just exist by default.

The best infrastructure is the kind you don’t think about.

Bun

Earlier this year, Bun shipped a native S3 client directly in the runtime.

Not an npm package. Not a third-party SDK. A built-in primitive.

import { s3 } from 'bun'
await s3.file('photo.jpg').write(imageData)

That one line replaces what used to be a whole process:

• Install the AWS SDK
• Configure credentials
• Create a client
• Build command objects

Fifteen lines of boilerplate to upload a file.

When a capability moves into the runtime, developers stop treating it as “infrastructure.” It stops being a decision. It’s just there.

Railway

Railway’s storage buckets went GA recently. Same idea.

The setup is trivial. Click a button. Name your bucket. Pick a region. Done. Credentials inject automatically.

Pricing is boring (in a good way):

• $0.015 per GB-month
• No API operation fees
• No egress fees¹

What you’re really buying isn’t storage. It’s the lack of friction.

• No IAM policies.
• No AWS console.
• No fear you misconfigured something subtle.

For most projects, that simplicity matters more than squeezing out a few cents.

When you combine a runtime that treats storage as first-class with a platform that removes setup entirely, the gap between idea and working system gets very small.

Databases

This enables something that would’ve sounded wrong a few years ago: not using a database at all.

For some workloads, you can store everything in object storage.

Write:

• one file with the content
• one small JSON file with metadata

Read it back by listing the bucket.

We’re trained to reach for Postgres immediately. But for append-only data, S3 works surprisingly well. Durable, cheap, scales automatically.

The downside is search. The moment you need complex queries, you’ll want a real database.

But if all you need is simple lists sorted by time, this is often enough.

What used to feel like a hack now feels reasonable—because the infrastructure underneath has become effortless.

Trend

This isn’t new.

Memory management used to be manual. HTTP clients used to be libraries.

Now garbage collectors handle memory, and fetch ships with the platform.

Each step feels small. Together, they change everything.

The surface area developers need to understand keeps shrinking.

When I say “infrastructure is disappearing,” I don’t mean it’s gone. I mean you don’t assemble it anymore. You consume it.

That’s a big deal for small teams.

Things that once required dedicated ops work now take minutes. Setup that used to take days happens during lunch.

The Anthropic acquisition of Bun suggests this trend will accelerate.² When a company depends on a runtime, it has every incentive to make that runtime boring, fast, and invisible. Bun staying open source means everyone benefits.

The stack is softening until it fades into the background.

That’s exactly where infrastructure should be.