Fireworks AI raises $1.5 billion at a $17.5 billion valuation as token volume hits 40 trillion a day

Fireworks AI has raised $1.505 billion in a Series D funding round at a $17.5 billion valuation, the company announced on July 15. The round was led by Atreides Management, Index Ventures, and TCV, with Nvidia among the participating investors — a notable detail given Fireworks' business is built on serving inference for other companies' AI workloads at scale.

The numbers behind the raise

Fireworks says its platform now processes more than 40 trillion tokens per day, up from 15 trillion a year earlier — a nearly threefold increase in raw usage. More significant than the volume figure is where that traffic is coming from: the company says over 95% of tokens processed now come from models fine-tuned or specialized on individual customers' proprietary data, rather than generic calls to off-the-shelf foundation models.

That shift matters commercially. Fireworks has crossed $1 billion in annualized revenue run rate, a figure the company says has grown fivefold over the past year. Charging for specialized, fine-tuned inference workloads generally commands better margins and stickier customer relationships than reselling access to commodity foundation model APIs — a distinction that likely explains why investors are willing to underwrite a $17.5 billion valuation for an infrastructure layer company rather than a model developer.

The investor list tells its own story

Beyond the three lead investors, the round included Lightspeed Venture Partners, Bessemer Venture Partners, Menlo Ventures, Insight Partners, Ontario Teachers' Pension Plan, Lone Pine Capital, and several smaller funds including 20VC and Operator Collective. Nvidia's participation is the most strategically interesting: the chipmaker has been selectively investing in AI infrastructure companies that drive demand for its hardware, and Fireworks explicitly says part of the new capital goes toward expanding compute infrastructure and deepening partnerships with cloud providers including Microsoft and Nvidia.

Why "specialized intelligence" is the pitch

Fireworks is positioning this round around a specific thesis: that the next phase of enterprise AI adoption isn't about renting access to ever-larger general-purpose models, but about building specialized systems trained on a company's own data and workflows. This is a deliberate contrast with the foundation-model providers — Fireworks doesn't build its own frontier models, it provides the infrastructure layer that lets enterprises deploy and serve fine-tuned versions of open models efficiently at scale.

This positioning places Fireworks alongside a small group of infrastructure-focused AI companies — rather than competing directly with OpenAI, Anthropic, or Google on model capability, they compete with cloud hyperscalers and other inference platforms on cost, latency, and customization tooling for running other people's models well.

What the funding will be used for

Fireworks has stated the capital will go toward three areas: expanding compute infrastructure to support continued token volume growth, growing its engineering headcount, and deepening its cloud provider partnerships. The 167% year-over-year token volume growth suggests the company is already capacity-constrained in some regions, making infrastructure expansion the most immediate priority.

What this signals for the broader market

A $17.5 billion valuation for an inference infrastructure company — as opposed to a model developer — reflects a maturing view among investors that enterprise AI value is increasingly captured at the deployment and customization layer, not just at the frontier-model layer. As more companies move from experimenting with off-the-shelf chatbots to running fine-tuned models on proprietary data, the infrastructure that makes that efficient and affordable becomes a durable business in its own right, independent of which foundation model happens to be winning any given month.