Fireworks AI is a cloud platform that lets companies build, fine-tune, and run AI models trained on their own proprietary data — without depending on expensive, closed frontier APIs from OpenAI or Anthropic. Customers choose from hundreds of open-source models (text, image, audio, multimodal), customize them via fine-tuning and reinforcement learning, then serve them through Fireworks' optimized inference stack. More than 95% of the 40+ trillion tokens processed daily come from customer-specialized models, not generic ones. The platform covers the full loop: model selection, evaluation (Eval Protocol), post-training, and high-throughput inference. Named customers span AI-native startups (Cursor, Harvey, Perplexity) and large enterprises (Uber, Shopify, DoorDash, Notion). Founded in 2022 by the PyTorch team from Meta, Fireworks frames its thesis as "specialized intelligence" — every company will eventually own its own tuned AI rather than rent a generic one. It now serves 10,000+ customers at $1B+ ARR, with a valuation of $17.5B after its July 2026 Series D.