Fireworks AI company profile: valuation, funding & revenue

This Fireworks AI company profile covers the company's valuation, revenue, funding rounds, investors and history. Fireworks AI works in AI infrastructure, with a focus on inference platform for generative AI. It was founded in 2022 and is headquartered in Redwood City, California, United States. Its latest valuation on record is US$17.5 billion (post-money) (July 2026). Every figure links to a public source, and anything the company hasn't disclosed is marked as such.

Revenue

Funding history

Fireworks AI timeline

Key dates from the sources on this page, oldest first. Metric snapshots are listed after the funding events.

• 2022 — Fireworks AI is founded in Redwood City by Lin Qiao (CEO), Dmytro Dzhulgakov, Benny Chen, Chenyu Zhao, Dmytro Ivchenko, James Reed and Pawel Garbacki.
• Oct 2025 — Series C: US$250M, valuing the company at US$4B. Investors: Lightspeed, Index Ventures, Evantic (co-leads); Sequoia Capital.
• Jul 2026 — Series D: US$1.505B, valuing the company at US$17.5B. Investors: Atreides, Index Ventures, TCV (leads); Nvidia, Lightspeed.
• May 2026 — Reported talks: Reports in 2026 described talks to raise at a valuation of about US$15 billion; not confirmed as closed.

Investors

Overview

Fireworks AI runs a platform that lets companies serve generative AI models quickly, reliably and at low cost. Developers send requests to Fireworks' API and get responses from open models such as Llama, DeepSeek, Qwen and Mistral, or from their own fine-tuned versions, without managing GPUs or the complex software needed to run models efficiently. Fireworks focuses on "inference", the work of running a trained model to answer requests, which has become the biggest and fastest-growing part of AI computing as applications move from experiments into daily use by millions of people.

Lin Qiao, the CEO, founded Fireworks in 2022 with a team of engineers who had built and run PyTorch, the open-source AI framework, at Meta. Qiao led Meta's PyTorch team, which supported AI workloads across Facebook, Instagram and WhatsApp at enormous scale. That experience in making AI software fast and dependable in production is the core of Fireworks' pitch. The company is based in Redwood City, California.

What Fireworks does

• Serverless inference: pay-per-token access to a large catalogue of open models, with no infrastructure to manage.
• On-demand and reserved deployments: dedicated GPU capacity for customers that need guaranteed speed and throughput.
• Fine-tuning and reinforcement learning: tools to adapt models to a company's own data and tasks, then serve the customised model on the same platform.
• Compound AI and agents: support for applications that chain several models and tools together, such as coding assistants and customer service agents.
• Performance engineering: a custom serving engine, quantisation and speculative decoding that reduce latency and cost.

Customers

Fireworks says it serves AI-native startups and large enterprises. Its publicly named customers have included coding tools, search and productivity companies, and businesses running high-volume AI features where speed matters, such as code completion that must respond in a fraction of a second. Many customers start with open models on Fireworks because the cost per request is lower than using a closed frontier model, then fine-tune those models as their products mature.

Business model

Fireworks earns money in two main ways: usage-based fees for tokens processed through its serverless API, and hourly or contracted fees for dedicated GPU deployments. Because inference demand grows every time a customer's product gains users, revenue can rise quickly without new sales. The flip side is that prices per token fall steadily across the industry, so Fireworks must keep lowering its own costs through better software and hardware use. The company hasn't published revenue figures, so we show revenue as not disclosed; press reports have described rapid growth in annualised revenue during 2025 and 2026.

Funding

Fireworks has been backed by Benchmark and Sequoia Capital, and by chipmakers Nvidia and AMD. In its Series C, announced in late 2025, it raised US$250 million at a US$4 billion post-money valuation. The round was co-led by Lightspeed Venture Partners, Index Ventures and Evantic, with Sequoia participating, bringing total funding to about US$327 million according to its law firm. Reports in 2026 said the company was in talks to raise at a valuation of about US$15 billion; we haven't seen that round confirmed, so the data above uses the Series C figure and lists the reported talks separately.

Competition

Fireworks competes with other inference platforms such as Together AI and Baseten, with specialist chip companies that offer their own clouds such as Groq and Cerebras, with GPU clouds such as CoreWeave and Lambda, and with the hyperscalers' AI services from Amazon, Microsoft and Google. Closed-model providers also compete for the same budgets by selling their own APIs. Fireworks' advantages are its team's depth in production AI systems, a strong focus on speed, and tools that make it easy to move from a stock open model to a customised one.

Fireworks AI at a glance

Fireworks AI is a Redwood City company founded in 2022 by Lin Qiao and former members of Meta's PyTorch team. It runs a platform for serving and fine-tuning open AI models. It raised a US$250 million Series C at a US$4 billion valuation in late 2025, co-led by Lightspeed, Index Ventures and Evantic, and was reported in 2026 to be in talks at a much higher valuation.

Why inference is the big prize in AI infrastructure

Training a frontier model is a large, one-off cost. Inference is ongoing: every chatbot reply, every code suggestion and every agent step requires computing power, and the total grows with the number of users and the complexity of tasks. Reasoning models, which "think" through problems step by step, and agents, which make many model calls to complete one job, have multiplied the amount of inference each user generates. Industry analysts expect inference to account for most AI computing spending over time. That shift has turned companies that make inference cheaper and faster into some of the most sought-after infrastructure investments.

Open models versus closed models

Closed models such as those from OpenAI, Anthropic and Google are accessed only through their makers' services. Open-weight models can be downloaded and run anywhere. For many tasks, a well-tuned open model is good enough and much cheaper, and companies gain control over data, latency and costs. The trade-off is operational complexity, which platforms like Fireworks remove. If open models keep closing the quality gap with the best closed models, demand for platforms that serve them should keep rising; if closed models pull far ahead, inference platforms may need to rely more on serving smaller, specialised workloads.

How inference platforms make models faster

Serving a large model efficiently involves many techniques. Quantisation stores model weights at lower precision so they use less memory and run faster with little loss of quality. Batching groups requests from many users so a GPU can process them together. Speculative decoding uses a small, fast model to guess the next few words and a large model to check them, speeding up responses. Caching reuses work for repeated parts of prompts, such as long system instructions. Fireworks and its competitors compete on how well they combine these methods, because small improvements in speed and cost per token add up to large differences at scale.

Questions for buyers

• What latency and throughput do you need, and can the provider guarantee them in writing?
• How do prices compare for your expected volume, including fine-tuned models?
• Where is data processed and stored, and is it used for training?
• How easily could you switch providers if prices or performance change?

Risks

• Price pressure: token prices keep falling, which squeezes margins for all inference providers.
• GPU supply and cost: access to the latest chips at reasonable prices is essential.
• Hyperscaler competition: large cloud providers can bundle inference with other services.
• Customer concentration: fast-growing AI startups can drive a large share of usage, and their fortunes can change quickly.

Latest developments (October 2026)

Fireworks AI raised a US$1.505 billion Series D in July 2026 at a US$17.5 billion post-money valuation, led by Atreides Management, Index Ventures and TCV, with Nvidia and Lightspeed participating, according to the company's legal adviser Orrick. Chief executive Lin Qiao has cited annualised revenue of about US$800 million by May 2026, up from about US$305 million at the end of 2025. This supersedes the October 2025 valuation described earlier on this page. See our AI cloud providers guide.

What to watch

Watch for confirmation of a new funding round, revenue milestones, large enterprise contracts, new hardware partnerships, and how Fireworks differentiates as inference platforms grow more alike.

Takeaways

Developers and enterprises can run open and customised models without building serving infrastructure. Founders can see how deep expertise from a big tech platform team becomes a startup advantage. Investors should weigh fast usage growth against falling prices per token across the industry.

Sources

Every figure on this page comes from the public sources below. Where sources disagreed, we used the company's own announcement or a primary news report. Private-company data changes; check the date beside each figure.

1. Orrick
2. AI Weekly
3. Orrick