3 min readnote
Opening ClockingCloud
A note on the compute we built for ourselves, now open to everyone
ClockingMinds
Why we built it
ClockingMinds exists to build a digital world that reflects reality. Reflection is how we do that for shopping and marketing, and it runs on image and video generation at the scale of one person at a time.
That work kept running into the same wall. GPUs were hard to get when we needed them and expensive when they sat idle. Every new model meant setting up drivers, containers, and autoscaling again before we could learn anything. Most of our time went into provisioning, not research.
So we built a thin layer for ourselves. It picks the right GPU for each job, keeps models warm, scales to zero when nothing is running, and bills by the second. It is tuned for one thing: generating pixels.
Why we are opening it
Talking with other teams, we heard the same story. Compute allocation is still one of the quietest bottlenecks in generative work. If the layer helped us, it would likely help them too.
Starting today, ClockingCloud (opens in a new tab) is open for anyone to use. One install, your model, and a GPU that is ready in seconds.
What does not change
Our focus is still Reflection. ClockingCloud is the infrastructure we already depend on, offered as is. Running it in the open makes it better for us too: every workload it serves sharpens the system that Reflection and Project Casper run on.