Last Updated: 16 July 2026
Fireworks, a startup that helps companies train and run AI models customized to their own data, has raised $1.505 billion in a Series D round. Co-founded by Lin Qiao, who previously led Meta's PyTorch team (the widely used open-source framework behind many of today's AI models), the company provides infrastructure for businesses like Uber, Shopify, and Revolut to adapt open AI models for their specific needs. The round values Fireworks at $17.5 billion.
The funding comes at a time when more enterprises are questioning whether renting general-purpose AI from a handful of providers is the right long-term strategy. Open-source AI models have gotten good enough that the performance difference is shrinking fast. According to Stanford's 2026 AI Index, the best closed model leads the best open model by just 3.3 percent. The broader market reflects that shift: the open-source AI model market is projected to grow from $19.05 billion in 2025 to $50.03 billion by 2030. For companies that want to keep sensitive data in house and avoid being locked into a single vendor, customized open models are becoming a practical alternative.
Atreides Management, Index Ventures, and TCV led the round. Existing investors Evantic, Lightspeed Venture Partners, and NVIDIA also participated, among others.
Many businesses that initially adopted a single large AI provider are rethinking that approach. Depending on one vendor means exposure to changing per-token prices, rate limits, and unpredictable model updates. That kind of dependency can be expensive to unwind. For companies in regulated industries or those handling sensitive customer data, open models offer more control over where data lives and how models behave. Fireworks reports that 95 percent of the tokens processed through its platform come from models that have been customized for a particular company's tasks, not used out of the box.
The company's revenue figures reflect that demand. Fireworks has passed $1 billion in annualized revenue run rate, a fivefold increase year over year from its last funding round. Daily token volume on the platform nearly tripled during the same period, growing from 15 trillion to more than 40 trillion. The platform now offers more than 200 models across text, image, and multimodal formats, with support for major new open-source releases within hours of launch.
"Fireworks has assembled one of the most elite and technical teams in AI, paired with technology that consistently sets the pace for the industry and commercial momentum that very few companies have ever achieved at this scale. We believe both frontier and open models will increasingly be used together. With a differentiated platform used and trusted by enterprises to train, customize, and serve open models in production, Fireworks is exceptionally well-positioned to unlock the value of specialized intelligence built on proprietary data and workflows."
Gavin Baker, CIO and Managing Partner at Atreides Management
The new capital will go toward hiring more engineers and expanding global compute capacity. Fireworks also plans to deepen partnerships with cloud providers, specifically naming Microsoft and NVIDIA as key partners going forward.
At its core, the company is betting that most businesses will eventually want to own their AI models, not rent them. That means scaling the tools that let an enterprise take an open-source base model and train it on its own data, workflows, and quality standards. Fireworks says the resulting models can match or outperform general-purpose closed models on specific tasks, while running faster and costing less. The company wants to handle the hard infrastructure work so that businesses without deep AI expertise can still build something tailored to their own operations.
Fireworks was founded by the team that built PyTorch, the open-source machine learning framework that became a standard tool for AI researchers worldwide. The seven co-founders bring experience from Meta and Google, with backgrounds in AI infrastructure, advertising systems, and large-scale machine learning. Lin Qiao, the CEO, previously led PyTorch at Meta. Other co-founders include Benny Chen, formerly a Meta ads infrastructure lead; Chenyu Zhao, who led Google's Vertex AI; and Dmytro Dzhulgakov, a core PyTorch maintainer.
What the platform actually does is fairly straightforward. Companies bring their data and pick from a library of open-source AI models, or bring models they already use, and then fine-tune them for specific tasks. Think of it as a workshop where a business can take a general-purpose AI engine and train it to understand its particular customers, products, or internal processes. The company keeps and controls that model. It can improve it over time.
"There are two paths forward for AI. In one, intelligence belongs to a few big labs, and everyone else rents it. In the other, every company in the world builds specialized intelligence of its own, shaped by the domain only it understands. We are building towards the second. Every company holds knowledge no one else has: its data, its workflows, its customers, its definition of quality. Fireworks turns that knowledge into specialized intelligence they own and can keep improving, giving every company the infrastructure to build, serve, and scale the models that matter most to their business."
Lin Qiao, Co-Founder and CEO of Fireworks
Customers include Doximity, Geico, Revolut, and Shopify.
The round brought together a wide mix of backers. Lead investors Atreides Management, Index Ventures, and TCV were joined by 20VC, Bessemer Venture Partners, Evantic Capital, Insight Partners, Lone Pine Capital, Lightspeed Venture Partners, Menlo Ventures, NVIDIA, Operator Collective, Ontario Teachers' Pension Plan, Original Capital, Prysm Capital, Quantum Capital, and TIME Ventures.
Earlier backers of Fireworks include Benchmark Capital, Sequoia Capital, and AMD. The investor base spans venture capital firms, growth equity funds, a pension fund, and corporate investors like NVIDIA. That variety suggests broad confidence across different types of institutional investors in what the company is building.
The $17.5 billion valuation is a notable figure for a company built around open models rather than proprietary frontier research.
