Nvidia-backed Fireworks hits $17.5 billion valuation as companies pursue cheaper AI models

The rising cost of advanced artificial intelligence models is prompting finance executives to explore more affordable open-source options. This shift is benefitting Fireworks, a cloud startup that competes with tech giants like Amazon and Google by offering models that developers can integrate into their applications. Recently, Fireworks announced that it has surpassed $1 billion in annualized revenue, a remarkable fivefold increase from the previous year. The Nvidia-backed company has also secured a substantial $1.5 billion funding round, pushing its valuation to $17.5 billion. Lin Qiao, the co-founder and CEO of Fireworks, expressed her enthusiasm during a CNBC interview, stating, "We're seeing super-linear demand. This is a once-in-a-lifetime opportunity to have this kind of market." While Fireworks is significantly smaller than industry leaders like Anthropic and OpenAI—each valued over $800 billion—it demonstrates that there is a growing dissatisfaction among companies regarding the current offerings from top labs. This development highlights that Amazon, Microsoft, and Google do not have an unchallenged grip on the cloud computing market. The increase in demand for user-friendly cloud infrastructure is evident, as companies like DigitalOcean have seen their shares rise by 149% this year. Additionally, CoreWeave, which rents Nvidia graphics processing units (GPUs), raised $1.5 billion in an IPO last year and is now valued at $42 billion. Fireworks operates in the inference cloud market, providing computing infrastructure for AI models. It is in competition with startups such as Baseten and Together AI, and is also supplying GPUs for training AI models in collaboration with firms like CoreWeave and Lambda. To expand its reach, Fireworks has formed alliances, including a partnership with Microsoft, enabling customers to access models through Fireworks while leveraging resources from over 20 suppliers, including Microsoft itself. This partnership allows Fireworks to support developers utilizing models from various sources, including Chinese firms like DeepSeek and MiniMax, as well as open-weight models from OpenAI. Qiao emphasized that Fireworks offers a unique value proposition by enabling clients to refine models with their own data, enhancing performance for specific tasks. Unlike competitors focused on generalized intelligence, Fireworks aims to deliver specialized intelligence. In a recent blog post, Microsoft CEO Satya Nadella echoed sentiments about the importance of retaining unique knowledge when using models, a point also made by Palantir CEO Alex Karp regarding the ownership of production means. Cost considerations play a critical role, as highlighted by Coinbase CEO Brian Armstrong, who noted the significant savings from using cheaper models. Founded in 2022 by Qiao and six co-founders, Fireworks currently employs around 200 people, with plans to expand to 600 by the end of 2026. The company has hired former Salesforce executive George Hu as president and aims to build a robust sales team. The newly acquired funds will also be used to enhance GPU resources and hire more technology experts. Fireworks is now processing a staggering 40 trillion AI tokens daily, a figure that underscores its growing significance in the AI landscape. By comparison, Google reported processing approximately 27 trillion tokens per day, while OpenAI's tools handled around 22 trillion. In a notable partnership, SpaceX has agreed to acquire Cursor, an AI coding startup that has become less reliant on major players like OpenAI and Anthropic, further showcasing Fireworks' expanding influence. Other notable clients include Elastic, GitLab, and MongoDB, with backing from investors such as Atreides Management, Index Ventures, TCV, and Nvidia.