Nvidia-backed Fireworks raises $1.5B at $17.5B valuation

Fireworks founders pose for a photo at the startup's headquarters in San Mateo, California. From left in front row: Chenyu Zhao, CEO Lin Qiao and Benny Chen. Back row: James Reed, Pawel Garbacki, Dmytro Ivchenko and Dmytro (Dima) Dzhulgakov. Source: Fireworks

Key Highlights

• Funding: Fireworks has raised $1.5 billion at a $17.5 billion valuation.
• Revenue milestone: The company is now generating over $1 billion in annualized revenue — five times more than last year.
• Backing: The startup is backed by Nvidia.
• Customer diversification: Fireworks once received over half of its revenue from coding startup Cursor but has since diversified as more companies adopt open models, according to CEO Lin Qiao.

Overview

The rising cost of the latest artificial intelligence models is increasingly breeding anxiety among finance executives, who have begun directing employees to consider open-source alternatives.

This trend is boosting cloud startup Fireworks, which competes with Amazon and Google to host models that developers can integrate into applications. The Nvidia-backed company announced Thursday that it has exceeded $1 billion in annualized revenue — five times its figure from last year — and has now closed a $1.5 billion funding round at a $17.5 billion valuation.

"We're seeing super-linear demand," Lin Qiao, Fireworks' co-founder and CEO, told CNBC in an interview at the company's headquarters in San Mateo, California. "This is a once-in-a-lifetime opportunity to have this kind of market."

Market Position

Fireworks is significantly smaller than Anthropic and OpenAI, which investors have valued above $800 billion each this year. It also trails the top names in technology, whose market capitalizations are measured in trillions. However, the startup's revenue milestone suggests that companies are not entirely satisfied with the models coming from the leading labs.

The achievement also provides fresh evidence that Amazon, Microsoft, and Google do not completely dominate the cloud computing market.

Broader Industry Trends

• DigitalOcean: Shares of the easy-to-use cloud infrastructure vendor are up 149% year-to-date as growth has accelerated.
• CoreWeave: The GPU rental company raised $1.5 billion in an initial public offering last year and is now valued at $42 billion.

Business Model and Alliances

By managing computing infrastructure for models, Fireworks operates in the inference cloud market alongside startups such as Baseten and Together AI. It has also begun providing GPUs for training AI models, similar to neoclouds CoreWeave, Lambda, and Nebius.

Rather than going it alone, Fireworks has started forming strategic partnerships:

• Microsoft Partnership (March 2026): Fireworks announced a partnership with Microsoft, which operates its own Foundry service for running open models. The arrangement allows Microsoft customers to access models through Fireworks, which draws computing power from more than 20 suppliers, including Microsoft.
• "Through Microsoft we can get much bigger reach," Qiao said.

Model Offerings

Fireworks provides developers with an easy way to adopt models from Chinese companies such as DeepSeek, MiniMax, and Z.ai. Open-weight models released by OpenAI last year are also available on the platform.

The concept, Qiao explained, is for clients to bring their own data — data that frontier labs do not have — and refine models until they achieve state-of-the-art performance for specific tasks.

"While Anthropic and OpenAI serve up 'generalized intelligence,' Fireworks can unlock 'specialized intelligence,'" she said.

Industry Context

The argument may sound familiar to those following the discourse of technology leaders.

Microsoft CEO Satya Nadella wrote in a Sunday blog post that "a company should be able to use a model without giving up the knowledge that makes it unique." Nadella was referencing remarks made by Palantir CEO Alex Karp on CNBC earlier this month.

Technical customers, Qiao noted, "want to know they own" their data and models.