Alumna of Fudan University Launches AI Cloud Service Company Valued at $4 Billion

Fireworks AI, an artificial intelligence cloud services startup, is in advanced talks for a new funding round that could value the company at $4 billion. Prominent venture capital firms, including Lightspeed Venture Partners and Index Ventures, are considering leading the round. If successful, the funding would mark a more than sevenfold increase in the company’s valuation over the past year, highlighting the strong interest from investors in the AI infrastructure sector, particularly in high-performance inference services. Founded in 2022 in Redwood City, California, Fireworks AI is led by Lin Qiao, its co-founder and CEO. Qiao earned her bachelor’s and master’s degrees from Fudan University before obtaining a Ph.D. in computer science from the University of California, Santa Barbara. Prior to founding Fireworks AI, she served as a senior engineering director at Meta, where she led a team of over 300 engineers responsible for developing and deploying AI frameworks, including PyTorch. PyTorch is one of the most widely used open-source machine learning frameworks globally, and her experience there provided her and her team with critical expertise in AI architecture and platform development. The company’s founding team is composed of six engineers who previously worked on PyTorch at Meta and one former Google AI engineer, giving Fireworks AI a strong technical foundation. “I have spent over 20 years in the tech industry, witnessing multiple waves of transformation—from cloud computing to mobile apps—but this AI revolution is a complete tectonic shift,” Qiao said in a media interview. She noted that while many companies want to adopt AI quickly, they often lack the infrastructure, resources, and talent to do so effectively. This gap inspired her to start Fireworks AI. Fireworks AI focuses on providing a platform that enables developers to run and customize open-source AI models more efficiently and at a lower cost. These models include DeepSeek, Alibaba’s Qwen, and Meta’s Llama series. Unlike traditional cloud giants like Amazon and Google, which offer all-in-one solutions, Fireworks AI operates as a more flexible inference service provider. Instead of directly purchasing and owning NVIDIA servers, the company aggregates server resources from various sources and sells access to these computing capabilities via an API. This approach allows it to offer optimized GPU usage through its proprietary FireAttention inference engine, significantly improving model processing speed and efficiency, and reducing customer costs. Qiao believes that as AI technology advances, both open-source and closed-source large language models will eventually converge in terms of quality and scale, due to the limited availability of public data and the growing similarity of mainstream architectures. In this scenario, customizing models with a company’s own data becomes essential for building a competitive edge. “Each company has a unique business model, goals, and operations. Fine-tuning is the key to leveraging AI, and that’s where Fireworks AI excels,” she explained. This focus on customization and optimization has attracted a growing number of clients, especially AI-native startups such as Cursor, an AI-powered coding assistant, and Perplexity, an AI-driven search engine. Strong demand from these clients has driven rapid revenue growth. According to reports, Fireworks AI’s annualized revenue has surpassed $200 million, with monthly revenue approaching $17 million. The company has projected that its annual revenue could reach $300 million by the end of the year. However, the fast-growing inference services sector is highly competitive. Fireworks AI faces challenges from rivals such as Together AI and Baseten, as well as from a new and formidable competitor—NVIDIA. In March, NVIDIA acquired Lepton, a inference service provider, and launched its own GPU cloud marketplace, moving from a hardware supplier to a direct service competitor. This development is expected to significantly impact the industry’s profit margins and competitive landscape. Despite the growth, Fireworks AI and similar companies still face financial challenges. According to sources, the company’s gross margin is approximately 50%, which is in line with industry peers but lower than the 70% or more typically seen in subscription-based software businesses. This is partly due to the need to maintain large amounts of idle server capacity to meet fluctuating demand, as well as competition from low-cost GPU cloud providers like CoreWeave. To improve profitability, Fireworks AI has told investors that its focus is on continuous GPU optimization to enhance resource efficiency and aims to raise its gross margin to 60%. Despite these challenges, the company remains attractive to investors. Before this potential $4 billion valuation round, Fireworks AI had raised about $77 million from top-tier venture capital firms like Sequoia Capital and Benchmark, as well as industry players such as NVIDIA, AMD, Databricks Ventures, and MongoDB Ventures. Some analysts see Fireworks AI as a potential acquisition target for large cloud providers seeking to offer more specialized AI services and reduce model deployment costs. Qiao remains confident in the company’s future. At a recent industry event, she said, “We bet on a large number of successful application developers. Those who can optimize models with their own data will stand out. We build tools and infrastructure to help them customize, integrate data, and improve the quality, speed, and concurrency of model inference. They will shine.”