How to Get Started With Compute in Under 5 Minutes
By Priya Nair, Product Analyst
Before you start
Good news: there's almost nothing to set up. Compute is designed so you can go from "curious" to "running" without an engineering project. Here's the exact path.
Step 1 — Open the door (zero cost)
Head to the Compute page and hit Start. There's no contract and no upfront cost — zero upfront · usage-based. This is the single biggest reason teams stall on AI adoption, and Compute removes it entirely.
Step 2 — Connect what you already use
Compute plugs into your existing stack with one-click integrations. No migration nightmares, no rip-and-replace. Most teams are live in their workflow in under two minutes.
Step 3 — Put it to work
Pick your first job. The highest-leverage starting points we see:
- Train and fine-tune large models
- Serve inference at scale
- Burst capacity for spiky workloads
- Replace fixed GPU clusters with elastic capacity
Step 4 — Lean on the features that move the needle
- Elastic Scaling. Go from one GPU to thousands and back in a click, automatically.
- Training & Inference. One platform for both heavy training runs and low-latency serving.
- Spot & Reserved. Blend spot and reserved capacity to cut costs without losing reliability.
- Zero Idle Spend. Pay only for active compute — no paying for hardware that sits cold.
Step 5 — Scale as you win
Because the model is a partnership, you scale at your own pace and Scale shares in the upside. Start small, prove value, then expand — the pricing rewards growth instead of punishing it.
Common questions
How fast can I scale up? From one to thousands of GPUs in a click — scaling is automatic and near-instant.
Do I pay for idle hardware? No. Compute is usage-based, so you only pay for active capacity.
That's it. Five minutes from now you could be running the same technology that helped build the AI industry.
Ready to see it for yourself? Start with Compute — free to start → Zero upfront cost. We only win when you win.