DeepInfra raises $107M Series B for purpose-built AI inference cloud
The AMW Read
Incremental raise in known inference infrastructure segment; significant as a segment-level signal of capital flowing to purpose-built inference clouds.
DeepInfra raises $107M Series B for purpose-built AI inference cloud
DeepInfra announced a $107 million Series B round co-led by 500 Global and angel investor Georges Harik, with participation from Nvidia, Supermicro, Samsung Next, and others. The company operates GPU clusters across eight U.S. data centers, processes nearly five trillion tokens per week, and reports 25x token growth since its Series A. It plans to expand its roughly 25-person engineering team and scale internationally.
Why it matters: This raise fits the "purpose-built inference layer" pattern, where startups differentiate from hyperscaler offerings by owning hardware end-to-end to guarantee predictable latency and cost for high-throughput agentic and open-source workloads. Nvidia's participation signals hardware vendors see inference infrastructure as a distinct market with demand for tight hardware-software integration. The round also reflects capital-cycle dynamics where inference-focused clouds attract serious investment as production AI demand shifts from training to serving.
Expert take: DeepInfra's rapid token volume growth confirms that inference infrastructure is becoming a critical bottleneck. However, the market remains contested: hyperscalers offer scale and integration, while spot GPU marketplaces provide flexibility. DeepInfra's ability to sustain ownership across eight data centers demands careful execution. The Series B provides runway to hire engineering talent and expand geographic footprint, but balancing capital intensity with competitive pricing will be key to long-term viability.