Fireworks AI

Fireworks AI in the enterprise

As of November 8, 2026. Every figure links to its source; company-reported numbers are labeled as such.

Fireworks AI sells computing for running and customizing open AI models. Businesses either pay per word processed on shared servers, rent dedicated chip capacity by the hour, or pay to train a version of an open model on their own data. Its pitch is that a tuned open model can match a closed model from OpenAI or Anthropic on a specific task at a fraction of the cost, and that the customer owns the result. Its best-known customers are software companies that build AI into their products, such as Cursor, Notion, GitLab, Harvey, Shopify and Uber. It is also available through Microsoft Foundry on Azure and through AWS, giving large companies a way to buy it from cloud providers they already use.

By the company's own figures, the business is large and growing fast. Annualized revenue passed $1 billion in July 2026, up from more than $280 million in October 2025, and daily volume rose to more than 40 trillion tokens (the small units of text that AI models read and write). Fireworks says more than 10,000 companies use it and that 95% of its volume comes from models customized for a customer. None of these numbers is audited. Outside analysts at Sacra estimate gross margin at about 50%, which is low for software and reflects the cost of renting Nvidia chips.

What to watch: whether new products for large enterprises, including Fireworks Sovereign (its software running in a customer's own data center), budget controls and audit logs, broaden its customer base beyond AI-native startups. Watch also how it handles price competition from the big clouds and rivals such as Together AI and Baseten, and whether customers stay after the October 2026 incident in which an intruder captured API keys stored in some customer accounts. The hiring of Meta's Dave Wehner as CFO and reported talks at a $30-40 billion valuation point to a company preparing for much larger scale.