DeepInfra

DeepInfra is a cloud platform built specifically for high-throughput AI inference, giving engineering and platform teams production access to 200+ open-source models through OpenAI-compatible APIs. It owns and operates its own GPU fleet across nine data centers and is SOC 2 and ISO 27001 certified with a zero data retention policy.

DeepInfra, founded in 2022 and headquartered in Palo Alto, California, operates an inference cloud engineered from the ground up for model serving rather than general-purpose compute. The platform exposes more than 200 open-source models — spanning text generation, embeddings, automatic speech recognition and text-to-image, including Llama, Mistral and Anthropic Claude offerings — through OpenAI-compatible APIs, so teams can migrate existing integrations with minimal code change. DeepInfra processes close to five trillion tokens weekly and says token volume has grown 25x since its Series A, driven by the continuous high-volume demand of agentic AI systems. Unlike resellers, DeepInfra owns and operates its GPU infrastructure: on 8 July 2026 it opened its ninth data center and first international site in Toronto, a 1.7 MW facility hosting over 1,000 NVIDIA Blackwell B300 GPUs, expanding low-latency capacity beyond its eight U.S. sites. CEO and co-founder Nikola Borisov framed the expansion around enterprises "moving from experimentation to production at unprecedented speed." The company closed a $107 million Series B on 4 May 2026, co-led by 500 Global and Georges Harik, with participation from NVIDIA, Samsung Next, Supermicro, Felicis, A.Capital Ventures, Crescent Cove, Peak6 and Upper90. Pricing is usage-based: language models bill per million input and output tokens (for example, Meta-Llama-3.1-8B-Instruct at $0.02/$0.04), most other models bill on inference execution time, and dedicated deployments are billed hourly per GPU (A100 80GB at $0.89, H100 80GB at $2.20, B200 180GB at $3.69). Three scheduling tiers — Flex at 0.8x, Standard at 1x and Priority at 1.5x base rates — let teams trade latency against cost. DeepInfra is SOC 2 and ISO 27001 certified with zero data retention, and collaborates with NVIDIA on Nemotron models and Blackwell deployment.

Category

Infrastructure & Cloud

Pricing

Usage-based

Target Market

CTOs, Platform Engineering Leaders, Enterprise Developers, Data Scientists, MLOps Engineers

Founded

2022

Headquarters

Palo Alto, United States

• ✓
  200+ open-source models on one API

  Serves text generation, embeddings, speech recognition and text-to-image models including Llama, Mistral and Claude through a single endpoint.

• ✓
  OpenAI-compatible APIs

  Existing OpenAI SDK integrations can point at DeepInfra with minimal code changes, lowering migration cost.

• ✓
  Owned GPU fleet across nine data centers

  DeepInfra owns and operates its own infrastructure, including a 1.7 MW Toronto site with 1,000+ NVIDIA Blackwell B300 GPUs opened July 2026.

• ✓
  Dedicated GPU deployments

  Custom LLM deployments bill hourly per GPU — A100 80GB at $0.89, H100 80GB at $2.20 and B200 180GB at $3.69 per GPU-hour.

• ✓
  Flex, Standard and Priority scheduling tiers

  Three service tiers priced at 0.8x, 1x and 1.5x base rates let teams trade latency guarantees against cost.

• ✓
  Enterprise security posture

  SOC 2 and ISO 27001 certified with a zero data retention policy on inference traffic.

• •
  Production LLM serving at scale

  Runs high-throughput text generation workloads that collectively process close to five trillion tokens per week.

• •
  Agentic AI backends

  Supplies the continuous, high-volume inference that autonomous agent systems generate in production.

• •
  Multimodal AI pipelines

  Provides embeddings, automatic speech recognition and text-to-image inference alongside text generation on one platform.

Language models bill per million input/output tokens; most other model types bill on inference execution time. Monthly invoicing thresholds span $20-$10,000 across five usage tiers. No free tier or trial is listed in the pricing documentation.

This page was written from 3 sources, 1 on domains other than deepinfra.com.

1. 1.deepinfra.com — deepinfra series bvendor
2. 2.globenewswire.com — DeepInfra Expands AI Inference Capacity with First Internati
3. 3.deepinfra.com — pricingvendor