Nvidia is rolling out a system of token credits for developers, offering free AI inference compute in a bid to lower the barrier to entry for building on its hardware.
What Nvidia is actually offering
Through the NVIDIA Developer Program and its NIM (NVIDIA Inference Microservices) API, developers can now access free inference credits for prototyping AI models. The setup is straightforward: 1,000 free credits on signup, expandable to 5,000 upon request, with no credit card required.
Those credits let developers run inference workloads on Nvidia’s Blackwell and Hopper GPUs. The API endpoints are OpenAI-compatible, meaning developers already building on OpenAI’s infrastructure can port their work over with minimal friction.
At GTC 2026 in March, CEO Jensen Huang leaned heavily into the concept of “AI tokens” as units of inference compute, suggesting that companies should be budgeting compute credits for their engineers the same way they budget salaries.
