Fireworks AI Hits $1 Billion in Revenue — And It Is Making Big Tech Look Slow

A startup founded by ex-Meta engineers just crossed $1 billion in annualized revenue, raised $1.5 billion in fresh funding, and is processing more AI tokens per day than Google and OpenAI combined in their developer tools.

Fireworks AI announced its Series D round on July 16, 2026, valuing the company at $17.5 billion. That is more than four times its $4 billion valuation from October 2025. Nine months. Four times the valuation. Five times the revenue.

For anyone paying attention to where AI infrastructure is actually heading — not where Amazon, Microsoft, and Google say it is heading — this is one of the most important data points of 2026.

What Is Fireworks AI?

Fireworks AI is an AI inference cloud platform. In plain language: it is the infrastructure layer that lets companies run AI models at scale without building and maintaining their own server farms.

The company was founded by engineers who previously worked on PyTorch at Meta — the open source framework that underlies a large portion of the AI models being trained and deployed today. That background shaped what Fireworks built: a platform optimized for speed, cost efficiency, and flexibility, with a particular focus on open source models rather than proprietary frontier systems like GPT or Claude Sonnet 5.

The core pitch is straightforward. If you are a company that wants to build an AI-powered product, you have two options. You can pay frontier model providers like OpenAI or Anthropic for API access — fast to start, expensive at scale, and you are dependent on their decisions about pricing and availability. Or you can use open source models like Llama, Mistral, or Qwen, run them on infrastructure built for the job, and keep control of your costs and your data.

Fireworks AI is the infrastructure for the second option. And increasingly, companies are choosing it.

The Numbers Behind the Story

The revenue trajectory is the clearest way to understand how fast this is moving.

Period	Annualized Revenue
Series B — July 2024	~$60 million
Series C — October 2025	$280 million
February 2026	$315 million
May 2026	$800 million
July 2026	$1 billion+

That is not steady growth. That is a near-vertical line. Revenue grew more than 400% year over year between early 2025 and early 2026, and then nearly tripled again between the Series C and the Series D just nine months later.

The customer base has grown at a similar pace. At the time of the Series B in July 2024, Fireworks had roughly 1,000 enterprise customers. By October 2025, that number had grown to more than 10,000 — a tenfold increase in roughly 15 months.

Named customers now include Uber, Shopify, Samsung, GitLab, MongoDB, Cursor, Harvey, Notion, DoorDash, Doximity, Elastic, and Upwork. These are not early-stage startups experimenting with AI. These are companies running AI at production scale, serving millions of users, where reliability and cost per token matter enormously.

40 Trillion Tokens Per Day — What That Number Actually Means

Fireworks AI CEO Lin Qiao said the platform now handles 40 trillion AI tokens per day. To understand why that number matters, it helps to compare it to what the largest AI labs are processing.

Company	Token Processing
Fireworks AI	40 trillion tokens/day
Google (developer tools)	~27 trillion tokens/day
OpenAI (developer tools)	~22 trillion tokens/day

A token is roughly three-quarters of a single word. Every prompt you send to an AI model, every response it generates, every document it processes — all of it is measured in tokens.

Fireworks AI, a startup with fewer than 300 employees at the time of its Series D, is processing more tokens per day than Google and OpenAI combined in their developer-facing infrastructure. That is the scale of what has been built here — and it has been built in roughly three years.

Who Backed This Round and Why It Matters

The Series D was led by Atreides Management, Index Ventures, and TCV. But the most significant name in the investor list is Nvidia.

Nvidia's participation in the Fireworks AI round is not a passive bet. Nvidia makes the GPUs that power almost every AI workload in the world. When Nvidia backs an AI infrastructure company, it is signaling that this company is part of the ecosystem it is building around its hardware. Fireworks pulling compute from more than 20 suppliers — and maintaining a close relationship with Nvidia — gives it flexibility in a market where GPU availability remains one of the key constraints on growth.

Other participants include Lightspeed Venture Partners, Evantic Capital, Bessemer Venture Partners, Menlo Ventures, Insight Partners, Ontario Teachers' Pension Plan, and Lone Pine Capital. The breadth of institutional capital here is notable — this is not a round led by one or two believers. It is a round that reflects a broad consensus that the AI inference cloud market is real, large, and not yet won by anyone.

Why Companies Are Choosing Fireworks Over AWS, Azure, and Google Cloud

This is the part of the story that matters most for the broader AI industry.

The conventional assumption heading into 2025 was that Amazon, Microsoft, and Google would capture the AI cloud market the same way they captured the general cloud market over the previous decade. They had the infrastructure. They had the enterprise relationships. They had the sales forces and the compliance certifications and the global data center footprints.

That assumption is proving incomplete.

The cost problem with frontier models

Running GPT-5 or Claude Opus at scale is expensive. For companies processing millions or billions of requests per month, the per-token cost of frontier model APIs compounds into a significant line item. Open source models — Llama, Mistral, Qwen, and others — can deliver comparable performance on many tasks at a fraction of the cost, provided you have the infrastructure to run them efficiently.

This cost pressure is visible even at the individual subscription level. Anthropic recently localized Claude pricing in India specifically to address cost sensitivity in price-conscious markets — a sign that even frontier model providers are feeling the pressure to compete on price.

Fireworks AI was built specifically to run open source models efficiently. Its inference engine is optimized for low latency and high throughput in a way that general-purpose cloud platforms are not. For the right use case, the cost difference is not marginal — it is an order of magnitude.

The control problem with Big Tech platforms

When a company runs its AI workloads on AWS Bedrock, Azure AI, or Google Vertex, it is doing so on infrastructure controlled by competitors. Amazon, Microsoft, and Google all have their own AI products. The same platform your company is using to run AI models also has a direct financial interest in the outputs of those models.

This data control concern extends beyond just cloud providers. As we covered in our piece on what AI knows about you from your online activity, companies and users alike are becoming more aware of what it means to hand data to platforms with competing interests. Fireworks AI is a pure-play infrastructure company. It does not have a competing AI product. Its only interest is in running your models as fast and cheaply as possible.

The speed problem

General-purpose cloud platforms are built to serve an enormous range of workloads. AI inference — particularly low-latency inference for real-time applications — is a specialized problem that benefits from specialized infrastructure. Fireworks was designed from the ground up for this use case, and the performance difference is measurable in the applications that depend on it.

CNBC noted that the company's growth "presents new evidence that Amazon, Microsoft and Google are not totally dominating in cloud computing." That is a carefully worded sentence, but the implication is significant.

The Cursor Dependency Is Gone

One risk that analysts flagged in Fireworks AI's earlier growth story was customer concentration. At one point, Cursor — the AI coding tool that has become one of the most popular developer products in the world — accounted for more than half of Fireworks' total revenue.

That concentration has been resolved. As the customer base has grown past 10,000 companies, Cursor has become a smaller percentage of total revenue even as it continues to scale. The revenue base is now distributed across coding tools, legal AI, enterprise search, conversational AI, agentic workflows, and consumer applications.

Cursor's success also reflects a broader shift in how developers work with AI. Tools like the OpenAI Codex Micro keyboard— a physical hardware device built around AI coding workflows — show how deeply AI has been embedded into the development process. Fireworks AI provides the backend infrastructure that makes these tools fast and cost-effective at scale.

This matters for two reasons. It reduces the platform risk that comes with any single customer accounting for a disproportionate share of revenue. And it confirms that Fireworks' value proposition extends well beyond the coding tool category where it first achieved scale.

What Fireworks Is Building Next

The $1.5 billion raised in this round will be deployed across three main areas.

Compute infrastructure expansion. Fireworks draws compute from more than 20 suppliers, and the new capital will be used to expand that footprint significantly. CEO Lin Qiao has said headcount will reach 600 by the end of 2026 — up from a much smaller team that built the platform to its current scale.

Sales and go-to-market. Fireworks has operated for most of its existence without a formal enterprise sales team. Customers largely found the product and signed up themselves. That model works at a certain scale and then hits a ceiling. The company hired former Salesforce executive George Hu as president in April 2026, and the new capital will fund the sales organization he is building.

Microsoft partnership deepening. In March 2026, Fireworks announced a partnership with Microsoft that allows Microsoft customers to access models through the Fireworks platform via Azure. Deepening that relationship — and potentially adding similar arrangements with other major cloud providers — is a core part of the expansion strategy.

What This Means for the AI Industry

The Fireworks AI Series D is a data point in a larger pattern that is becoming clearer in 2026.

AI infrastructure is not a winner-take-all market. The assumption that Big Tech would consolidate this space the way it consolidated general cloud computing is being challenged by the actual behavior of enterprise customers. Companies are making deliberate choices to run AI on specialized infrastructure, and they are making those choices for reasons that have nothing to do with novelty or experimentation — cost, control, and performance are the deciding factors.

The open source model ecosystem is also more capable than the frontier model narrative suggests. Even Linus Torvalds — the creator of Linux — recently declared that open source development is no longer anti-AI, telling critics to "fork it or walk away." That shift in the open source community's relationship with AI tools is accelerating the adoption of AI in production environments — exactly the workloads Fireworks is built to serve.

This is the bet Fireworks was built on. At $1 billion in revenue and $17.5 billion in valuation, that bet is looking correct.

What Frontier Model Labs Are Watching

For companies like Anthropic and OpenAI, the Fireworks AI story raises a genuine strategic question.

Both companies have built their businesses around API access to proprietary frontier models. That model works when the performance gap between frontier and open source is large enough to justify the price premium. As open source models close that gap — and as inference infrastructure like Fireworks makes them easier and cheaper to deploy — the calculus changes.

The question of data privacy also plays into this dynamic. Enterprises increasingly want to know exactly where their data goes and who has access to it. As our piece on how Meta used Instagram photos for AI without user consent showed, the backlash against opaque data practices is real and growing. Fireworks AI's model — where enterprise data stays within the customer's own deployment — addresses that concern directly.

Fireworks AI's $1 billion revenue is partly a bet that this compression is coming — and that the companies best positioned to benefit are the ones building infrastructure for whatever models enterprises decide to run.

Frequently Asked Questions