Welcome to the fourth issue of Worth Noting, a newsletter from Notable Capital where we share the ideas, conversations, and frameworks that founders and operators are actually paying attention to.

For a long time, cloud providers like AWS, Azure, and GCP owned the compute and were the undisputed rulers of the internet.

They built and owned the roads, while infrastructure software vendors (think MongoDB, Redis, or Gitlab) collected a small toll at one on-ramp. That toll (called the attach rate) ran about 5–10% of what customers spent on cloud. It supported decent margins, but it also meant that the overwhelming majority of the roughly $200 billion in annual cloud spending flowed to Amazon, Microsoft, and Google, while the software vendors sitting on top fought for a sliver of the bill.

And the toll collectors were always one decision away from getting cut out entirely. The road owner could build another on-ramp, slash the toll, or reroute traffic overnight. As AI drives the cost of writing code toward zero, that's exactly what started happening. The already-small attach rate drops even lower.

The most valuable infrastructure companies of this era looked at that dynamic and decided to stop collecting tolls and start building their own roads. Dan Cahana (an investor at Notable) has been calling them workload clouds, and their fingerprints are already on the infrastructure decisions AI agents are making right now.

Snowflake was the pioneer here. Instead of selling data warehouse software atop the cloud providers, with customers running on their own infrastructure, they took over the whole workload. And now a new generation of companies is running this same playbook across every major compute workload.

Vercel built the frontend cloud. fal built the generative media cloud. Browserbase built the browser cloud. Supabase and Neon built the Postgres cloud. Railway built the Python cloud.

Notably, these often started as niche markets that cloud providers weren’t paying close attention to and investors had written off as too small, because attach rates looked so low.

Here’s the playbook they’re running: 1) Pick a workload that's painful for developers to run themselves, 2) offer a service instead of software, and 3) get exceptionally good at running it.

Taking on the big cloud providers means competing where they’re weakest, which is speed and developer experience. From there, you optimize relentlessly until you can run that workload better and cheaper than any individual customer could on their own.

If owning the workload is such an advantage, why hasn’t everyone tried this approach? Because in the early days, owning the workload is much harder than shipping software — margins are lower, enterprise customers are out of reach until you've earned credibility with developers. But over time, these businesses become more defensible than anything built on a software attach rate.

You can screenshot a UI and rebuild it. You can't replicate the distributed systems work, the hardware optimization, the reliability engineering that accumulates from years of running a specific workload at scale. That expertise quietly compounds in the background, invisible to customers but increasingly difficult for competitors to close the gap.

Today, agents make more infrastructure decisions autonomously. When Claude Code builds something in JavaScript, it deploys to Vercel by default. When it builds in Python, it deploys to Railway. There’s a ridiculous advantage to being the default, and there’s never been a better time to build a workload cloud.

This piece is adapted from Dan Cahana's original post, Workload Clouds and the Death of the Attach Rate. Watch his video explainer below:

In this episode of Notable Perspectives, we sit down with Matt Jahansouz, Chief People Officer of Quince. Quince has crossed $1 billion in revenue and a $10.1 billion valuation by radically rethinking the end-to-end consumer and e-commerce model.

Matt unpacks the people leadership playbook behind that growth: how to close elite talent in a bidding war, why workforce planning past 90 days is largely wasted time, and what it actually means to build a team for an AI-driven world.

Listen now: YouTube, Spotify, Apple