Credits-Based Billing Tied to GPU Costs

This pricing model turns GPU time into billable usage instead of hoping a flat seat fee covers unpredictable generation demand. A creator who needs a few worlds each month can stay on a lower plan, while a studio rushing a pitch deck, game prototype, or robotics sim can buy extra credits for a short burst of heavy rendering. That keeps spend tied to costly actions like generation and mesh export, which is critical because each output consumes real accelerator capacity.

• Marble charges by credits for concrete compute events, not vague platform access. World generation, draft generation, and high quality mesh export each draw separate credit amounts, so revenue rises with the exact workflows that burn the most GPU time.
  1 worldlabs
• This is closer to cloud infrastructure billing than classic design software. In mature SaaS, one extra click costs almost nothing. Here, a larger or higher fidelity scene means more model inference and more serving cost, so top ups protect margins when customers suddenly spike usage.
  2 worldlabs 3 sacra
• The same cost logic shapes competition. NVIDIA is pushing open Cosmos models into robotics and simulation, and Meta is advancing open world models for planning. That puts pressure on standalone model pricing, so World Labs needs to capture value through workflow, editing, exports, and enterprise integrations, not raw generation alone.
  3 sacra 4 nvidia 5 meta

The next phase is a race to widen the gap between what users pay and what each generated world costs to serve. If World Labs improves inference efficiency with LLM style serving tricks, it can lower effective compute cost, support bigger usage bursts, and keep more of each extra credit dollar as gross profit.