DeepInfra raises $107M to scale AI inference infrastructure

What's the deal? DeepInfra, a Palo Alto-based cloud platform built specifically for AI inference, has closed a $107M Series B round. The company processes nearly five trillion tokens per week and serves enterprises running open-source and agent-driven AI workloads.

The round was co-led by 500 Global and Georges HarikDealroom has a profile for this one. Try Dealroom →, one of Google's earliest engineers. NVIDIA, Felicis, Samsung Next, Supermicro, and several other investors also participated.

DeepInfra was built by the team behind imo, a messenger app that scaled to more than 200 million users globally.

Why now? The AI industry is shifting from training models to running them in production — and that's where inference comes in. As agentic AI systems become the norm, each user request can trigger dozens of model calls, creating continuous, high-volume demand for compute.

Open-source models are rapidly closing the gap with proprietary ones, making inference infrastructure even more critical. Enterprises want to avoid vendor lock-in and control costs while still deploying powerful AI. Deloitte estimates inference could account for roughly two-thirds of all AI compute this year.

"Inference is no longer a thin layer — it's the system constraint that will define the majority of workloads," said Nikola Borisov, co-founder and chief executive officer of DeepInfra. "Most cloud platforms weren't built for this always-on, distributed model."

What could go wrong? DeepInfra is competing against hyperscalers like AWS, Google Cloud, and Azure — all of which are investing heavily in inference capabilities. Purpose-built platforms must prove they can offer meaningfully better economics and performance to win enterprise workloads away from incumbents.

The company also faces a moving target: as model architectures evolve and hardware accelerators change, inference infrastructure must adapt quickly or risk obsolescence.

The signal: NVIDIA's participation as a corporate investor underscores how chipmakers are moving downstream, backing the infrastructure layer that drives demand for their own hardware. With 500 Global, Felicis, and Samsung Next all converging on the same bet, the round signals growing investor consensus that purpose-built inference platforms — not general-purpose clouds — will capture the economics of always-on, agentic AI workloads.

Read more: globenewswire.com