A data scientist opens a laptop in a coffee shop. The notebook is innocent enough - a few lines of pandas, a Dask import, a dataset that happens to be 250 terabytes. She runs one cell. Somewhere in a cloud region she will never see, a thousand virtual machines wake up, do the work, report the cost in dollars, and switch themselves off. She never opened a Docker file. She never typed the word Kubernetes. That quiet trick is Coiled.
This is the company as it stands now: small, profitable in attention if not yet in headlines, and oddly central to a slice of computing most people never think about. Coiled does not sell flash. It sells the absence of friction. The product is, more or less, the disappearance of the most annoying afternoon in a data scientist's week.
"Quite literally burst to the cloud from your laptop."
Python won. The cloud stayed hard.
Python ate data science. It did not, however, learn to share. The language that millions of analysts and researchers use to think runs beautifully on a single machine and then hits a wall the moment the data grows past memory. The standard answer - rewrite everything in a distributed framework, learn container orchestration, become a part-time DevOps engineer - is the kind of answer that makes people quietly give up and just sample their data instead.
The gap is not a small one. On one side: a scientist who knows pandas. On the other: a cluster that speaks YAML. Bridging it has historically meant hiring an infrastructure team, or surrendering to a heavyweight platform that wants to own your data, your bill, and your weekends. Most teams chose to shrink the problem rather than scale the tools.
"Set up is half the battle with the cloud. Coiled made this easy."
A free library, then a company
Back in 2015, Matthew Rocklin wrote the first version of Dask - an open-source library that lets familiar Python tools like pandas, NumPy, and scikit-learn run across many machines. Millions of developers adopted it. It was, by any measure, a success. It also made almost no money, which is the traditional fate of beloved open-source projects.
In 2020 Rocklin made the bet: the hard part was never the math, it was the plumbing. Spun out of Anaconda, Coiled was built to be the commercial sequel to Dask - not a rewrite, not a lock-in, but the boring, reliable machinery that gets a Dask cluster running in someone's own cloud account in under a minute. He was joined by early collaborators including Hugo Bowne-Anderson and Rami Chowdhury. The pitch to investors was almost contrarian: we will not own your data, and we will help you spend less.
"Dask for everyone, everywhere."
The Coiled timeline
One decorator, the whole cloud
Coiled's central magic is environment replication without Docker. It scans your local Python setup, ships it to the cloud, and runs your code there - in your own AWS, GCP, or Azure account. Your data never leaves your walls. The product has since grown past Dask clusters into three shapes: on-demand clusters, serverless functions you summon with a single decorator, and SLURM-style batch jobs for the people who think in arrays.
There is a web UI for debugging, spot-instance handling to keep the bill down, GPU support for the ambitious, and - the detail that data engineers actually love - a dollar figure attached to every job. The cloud, finally, with a price tag you can read.
Distributed Python in your cloud, autoscaling, auto-shutdown.
Run pandas, Polars, or DuckDB on a VM with one decorator.
SLURM-style job arrays with per-job cost tracking.
No Docker, no Kubernetes, no YAML to maintain.
Who's actually running on it
The customer list reads like a tour of problems too big for a laptop: NASA and climate teams doing geospatial work, NVIDIA on accelerated computing, Moderna in genomics, the US Air Force, the quants at D.E. Shaw, and KoBold Metals hunting for minerals with machine learning. Floodbase models floods. UrbanFootprint maps cities. The common thread is data measured in terabytes and patience measured in minutes.
Funding, by the round
"Processing a 250 TB dataset with Coiled, Dask, and Xarray."
Make the cloud boring
Coiled's stated goal is almost modest: let Python people scale their existing workflows without rewriting code or learning infrastructure. No new language, no migration project, no platform that quietly becomes the most expensive line item in the budget. The business model is the same shape as the philosophy - usage-based, runs in your account, charges by the CPU-hour, and leaves the open-source Dask free for anyone who never wants to pay at all.
It is a bet that the winning move in infrastructure is to be invisible. The best plumbing is the kind nobody notices. Coiled would like to be the company you forget you're using.
Why it matters next
AI and data workloads are only getting heavier, and the people doing the work are increasingly not infrastructure specialists. They are scientists, analysts, and researchers who think in Python and would prefer to keep it that way. Every gigabyte that grows into a terabyte is another moment where someone has to choose: sample the data and lose the signal, or scale the tools and keep it. Coiled exists to make the second choice the easy one.
Which brings us back to the coffee shop. The same scientist, the same laptop, the same 250 terabytes. Only now the wall isn't there. She runs the cell, the thousand machines wake and sleep, the answer arrives, and she goes back to her coffee - never having learned a single thing about the cloud that did all the work. That, in the end, is the whole point.