DeepInfra Closes $107M Series B — Processes Nearly 5 Trillion Tokens/Week, Revenue Tripled Since January 2026
DeepInfra·medium signal
DeepInfra raised $107M co-led by 500 Global and Georges Harik, with strategic investment from NVIDIA and Supermicro, to scale its purpose-built inference cloud. Token volume grew 25x since Series A, now processing nearly 5 trillion tokens per week across 190+ open-source models. Revenue tripled since January 2026. For builders running multi-agent systems, inference cost and throughput are the binding constraints — DeepInfra's growth confirms that infrastructure is the picks-and-shovels play of the agent era.