DeepInfra funding, valuation and investors

Overview

DeepInfra is a Palo Alto-based cloud inference platform that lets developers and enterprises run open-source and proprietary LLMs via OpenAI-compatible APIs without owning GPU fleets. Founded in 2022 by Nikola Borisov and team (ex-imo messenger, 200M+ users), the company owns and operates its GPU infrastructure, supports 190+ models, and holds SOC 2 and ISO 27001 certifications with zero-data-retention options. In May 2026 it raised $107M in Series B co-led by 500 Global and Georges Harik, with NVIDIA, Felicis, Samsung Next, and Supermicro participating. The company also partners with Nvidia on Nemotron model serving.

Why DeepInfra is interesting

DeepInfra's $107M Series B in May 2026 came as the platform processed nearly five trillion tokens per week — proof that inference economics, not training clusters, is where recurring AI infrastructure revenue concentrates.

Product & use cases

DeepInfra provides a fully managed inference cloud with pay-per-token pricing, OpenAI-compatible endpoints, and enterprise security (zero data retention, compliance certifications) for running open and proprietary models in production.

• Startups and scaleups deploying open-weight LLMs without capital expense on GPU clusters
• Enterprises routing inference workloads to cost-optimized providers with compliance requirements
• Agentic AI applications needing high-throughput, low-latency model serving at scale

Key facts

• Series B (May 2026): $107M co-led by 500 Global and Georges Harik (GlobeNewswire)
• Processes nearly 5 trillion tokens per week on owned GPU infrastructure
• 190+ open-source models; OpenAI-compatible API; SOC 2 and ISO 27001 certified
• Series A (Apr 2025): $18M; team built imo messenger scaling to 200M+ users
• Nvidia partnership for Nemotron inference; Supermicro hardware investor

Funding history (newest first)

Competitive landscape

Edge: DeepInfra combines owned infrastructure with imo-era scaling expertise — processing trillions of tokens weekly while undercutting some GPU-cloud incumbents on per-token economics for supported open models.

Inference is the recurring cost center for most AI products as open models proliferate. DeepInfra competes in a crowded layer where margins depend on GPU supply, utilization, and routing — differentiation comes from model catalog breadth, compliance, and cents-per-token pricing rather than proprietary silicon.

• Together AI direct

  Inference and fine-tuning platform for open models; similar developer audience.

• Fireworks AI direct

  Fast inference for open and custom models; competes on latency and price.

• Baseten direct

  Model serving and deployment platform; stronger on custom model ops.

• AWS Bedrock / Azure OpenAI incumbent

  Hyperscaler bundled model APIs; DeepInfra wins on open-model breadth and cost for some workloads.

Notable stories

• DeepInfra's founding team built imo, a messenger app with 200M+ users — bringing hyperscale consumer infra experience to inference economics (GlobeNewswire, May 2026).
• Georges Harik, one of Google's earliest engineers, co-led the Series B — a strategic signal for production-grade inference at scale.

Industries

Market / IPO context

Related funding articles

Venture Capital Tracker pieces that cover DeepInfra's financing or category context.

• DeepInfra's $107M Series B: Cheaper, Faster Model Inference at Scale