DeepInfra Expands Startup Support Through NVIDIA Inception Offering

New benefit builds on DeepInfra’s expanded DeepStart program for early-stage AI companies

DeepInfra, a purpose-built cloud platform for high-throughput AI inference, announced a new benefit for members of the NVIDIA Inception program, providing eligible startups with 10 billion cached-input tokens for the NVIDIA Nemotron family models on DeepInfra’s inference platform. At the same time, the company is also expanding DeepStart, its program for early-stage AI companies, to provide 50 billion free tokens.

“AI startups are moving from prototype to production faster than ever, and infrastructure can quickly become the constraint,” said Nikola Borisov, co-founder and CEO of DeepInfra. “We want teams to be able to focus on building products instead of managing infrastructure. By expanding DeepStart and working with NVIDIA Inception, we’re giving more companies the capacity to test, build and scale on the same infrastructure they can use in production.”

The new NVIDIA Inception benefit gives startups access to NVIDIA Nemotron family models through DeepInfra’s serverless inference platform, with no infrastructure to provision or manage and production-grade throughput on NVIDIA’s latest-generation accelerated computing infrastructure. NVIDIA Inception members accepted into the program can also take advantage of DeepStart, giving them additional resources to use across models available on DeepInfra.

Also Read: AiThority Interview with Gou Rao, co-founder and CEO at NeuBird AI

Supporting AI Startups From Prototype to Production

DeepInfra created DeepStart to give early-stage AI companies access to production-grade inference infrastructure as they develop and scale new applications. The expanded program now provides qualifying startups with approximately 50 billion free tokens, up from the program’s previous offering of 1 billion free tokens. Credits are available for one year following approval.

The programs come as AI applications increasingly move into production, where inference cost, latency, and reliability become critical constraints. DeepInfra’s vertically integrated platform combines dedicated AI compute, optimized software, and a fully managed developer experience, allowing companies to run more than 200 models through OpenAI-compatible APIs without having to build or manage their own GPU infrastructure.

DeepInfra works closely with NVIDIA across its infrastructure and model ecosystem: it has first-party integration with NVIDIA OpenShell, and DeepInfra was part of a select group of collaborators to run an independent benchmark on NVIDIA Vera CPU.

Companies including OpenCode, LiveKit and humans& have used DeepStart to build and scale AI products on DeepInfra.