Fireworks raises $1.5 billion at $17.5 billion valuation

Fireworks has raised $1.5 billion in new funding at a $17.5 billion valuation, the Nvidia-backed artificial intelligence cloud company said Thursday. The San Mateo, California-based startup said annualized revenue has passed $1 billion, roughly five times its level a year earlier, as software developers and corporate customers use more open-source and open-weight AI models.

The financing was led by Atreides Management, Index Ventures and TCV, according to the company. Nvidia also joined the round, alongside Evantic and Lightspeed Venture Partners.

Fireworks runs cloud infrastructure that lets developers place AI models inside applications. The company competes with large cloud providers including Amazon and Google in hosting models, and it operates in the inference cloud market, where providers handle the computing needed to generate responses from AI systems after they have been trained.

Chief executive Lin Qiao told CNBC in an interview at Fireworks’ headquarters that demand is rising faster than expected. “We’re seeing super-linear demand,” Qiao said. “This is a once-in-a-lifetime opportunity to have this kind of market.”

Demand shifts toward cheaper AI capacity

Fireworks’ growth comes as finance executives scrutinize the cost of advanced AI models and direct teams to consider open-source alternatives, CNBC reported. Qiao said Fireworks’ cost is “five to 10 times cheaper” than closed models of equivalent quality.

The company provides access to models from Chinese companies including DeepSeek, MiniMax and Z.ai. It also offers open-weight models released by OpenAI last year. Qiao said customers can use their own data to refine models for specific business tasks, describing Fireworks’ approach as enabling “specialized intelligence,” compared with the “generalized intelligence” offered by Anthropic and OpenAI.

Microsoft has also become a distribution partner. Fireworks announced an arrangement in March with Microsoft Foundry, Microsoft’s service for running open models. Qiao told CNBC the partnership gives Fireworks a wider route to customers, while Fireworks draws computing power from more than 20 suppliers, including Microsoft.

The company has expanded beyond inference into providing graphics processing units for training AI models, a field that includes CoreWeave, Lambda and Nebius. CoreWeave, which rents Nvidia GPUs, raised $1.5 billion in an initial public offering last year and is now valued at $42 billion, according to CNBC.

Customer base broadens

Fireworks was founded in 2022 by Qiao, a former Meta director, and six co-founders. The company has about 200 employees, and Qiao told CNBC she expects staffing to reach 600 by the end of 2026.

Fireworks hired former Salesforce executive George Hu as president in April. The company plans to build a larger sales organization after relying for years on customers signing up themselves, and it expects to use the new capital to secure more GPUs and recruit technical staff.

Qiao said Fireworks now processes 40 trillion AI tokens a day. CNBC compared that with Google’s May disclosure that its AI models were processing about 19 billion tokens per minute for developers, or more than 27 trillion a day, and OpenAI’s March statement that its developer tools were processing 15 billion tokens per minute, or about 22 trillion a day. A token is roughly three-quarters of a word.

Last year, about half of Fireworks’ revenue came from AI coding company Cursor, CNBC reported. Qiao said the business is now more diversified. Fireworks customers include Elastic, GitLab and MongoDB, while CNBC reported that SpaceX agreed in June to acquire Cursor in a $60 billion stock transaction expected to close this quarter.

This story draws on original reporting from CNBC.