Fireworks founders pose for a photo at the startup’s headquarters in San Mateo, California. Front row from left: Chenyu Zhao, CEO Lin Qiao, and Benny Chen. Back row: James Reid, Pavel Garbakki, Dmytro Ivchenko, Dmytro (Dima) Djurgakov.
firework
The cost of modern artificial intelligence models is causing growing anxiety among financial executives, who are starting to tell their employees to consider open source alternatives.
That’s what’s driving cloud startup Fireworks. Amazon and google Hosts models that developers can incorporate into their applications. of NvidiaThe company, backed by , on Thursday announced annual revenue of more than $1 billion, five times as much as last year, and raised $1.5 billion at a valuation of $17.5 billion.
“We’re seeing very linear demand,” Fireworks co-founder and CEO Lin Qiao told CNBC in an interview at the company’s headquarters in San Mateo, California. “It’s a once-in-a-lifetime opportunity to have a market like this open.”
fireworks are smaller human and OpenAIInvestors this year valued it at more than $800 billion annually, dwarfing even the tech industry’s top companies with market capitalizations in the trillions. But the startup’s revenue milestones suggest the company isn’t entirely satisfied with the model coming out of its top lab.
As a result of this achievement, Amazon microsoft Google doesn’t completely dominate cloud computing.
Share of easy-to-use cloud infrastructure vendors digital ocean Growth has accelerated, increasing by 149% so far this year. coreweaveThe company, which rents out Nvidia graphics processing units (GPUs), raised $1.5 billion in an initial public offering last year and is now worth $42 billion.
Fireworks does business in the inference cloud market alongside startups like Baseten and Together AI by managing the computing infrastructure for models. In addition, Neocloud CoreWeave, Lambda, Nevius.
Rather than going it alone, Fireworks is starting to form partnerships.
In March, the company announced a partnership with Microsoft, which will roll out its own Foundry service to run an open model. The deal will make the model available to Windows and Office customers through Fireworks, which relies on computing power from more than 20 suppliers, including Microsoft.
“Through Microsoft, we have even greater reach,” Qiao said.
Fireworks allows developers to easily adopt models from Chinese companies such as DeepSeek. mini max and Zai. We also have the OpenAI open weight model released last year. The idea is for customers to bring in their own data that Frontier Labs doesn’t have and refine the model until it provides cutting-edge performance for a specific task, Qiao said.
While Anthropic and OpenAI provide “generalized intelligence,” Fireworks can unlock “specialized intelligence,” she said.
This argument may be familiar to those who follow the debates of some of the biggest names in the technology world.
“Companies should be able to use models without giving up the knowledge that makes them unique,” Microsoft CEO Satya Nadella said in a blog post on Sunday.
What Nadella said Palantir CEO Alex Karp made this statement on CNBC earlier this month.
Technology customers “want to know they own the means of production,” Karp said. “It will not be transferred to anyone else.”
Dollars and cents are also a factor. virtual currency exchanger coinbase CEO Brian Armstrong wrote in a June 1 post that the company uses cheaper models when it makes sense.
“Compared to similar quality closed models, our cost is 5 to 10 times lower,” Qiao said.
Former meta Director Qiao and six co-founders founded Fireworks in 2022. The company has approximately 200 employees. Qiao expects the number of employees to reach 600 by the end of 2026.
“This year is the year we really step on the gas,” Qiao said.
Fireworks supplier sales force Executive George Hu will become president in April. The startup plans to build a strong sales team after years of getting customers to sign up on their own. The new funding will also help Fireworks acquire more GPUs and hire more engineers.
Developers increasingly rely on Fireworks to handle requests.
According to Chao, Fireworks currently processes 40 trillion AI tokens per day. google revealed in May that its AI models were processing approximately 19 billion tokens per minute for developers, which means more than 27 trillion per day. OpenAI announced in March that its developer tools were processing 15 billion tokens per minute. This translates to approximately 22 trillion tokens processed per day. Each token corresponds to about three-quarters of a word.
As of last year, about half of Fireworks’ revenue came from AI coding startup Cursor, which reduced its reliance on OpenAI and Anthropic and built custom models called Composer.
“We are much more diverse now,” Qiao said. In June, Elon Musk space x has agreed to acquire Cursor in a $60 billion stock deal, with the deal expected to close this quarter. Other Fireworks clients include: elastic, GitLab and Mongo DB.
Atreides Management, Index Ventures, and TCV led Fireworks’ new round. Nvidia also participated, as did Evantic and Lightspeed Venture Partners.
Spotlight: OpenAI CEO: China’s open source model is getting much better
