Funding

Fireworks AI Raises $1.5 Billion at a $17.5 Billion Valuation as Its Inference Business Hits $1B Run Rate

Newaiera Desk · 2026-07-23 · 4 min read

Fireworks AI closed a $1.505 billion Series D at a $17.5 billion valuation — more than 4x its price ten months ago — after its inference platform crossed $

While the frontier labs fight over who has the smartest model, a quieter business has been compounding underneath them: running everyone else's models, cheaply and at scale. Fireworks AI just put a number on how big that business has gotten.

The company announced a $1.505 billion Series D at a $17.5 billion valuation, led by Atreides Management, Index Ventures and TCV, with participation from Nvidia, Lightspeed Venture Partners, Evantic Capital, Bessemer Venture Partners, Menlo Ventures, Insight Partners, Ontario Teachers' Pension Plan and Lone Pine Capital.

The number that justifies it

Fireworks sells inference infrastructure — the compute layer that actually runs AI models in production, as opposed to training them. That business has just crossed $1 billion in annualized revenue, up roughly fivefold from a year earlier. Daily token volume on the platform has climbed to more than 40 trillion tokens a day, up from 15 trillion over the same period.

For context on the trajectory: Fireworks last raised money in October, a $250 million round at a $4 billion valuation. Ten months later, the valuation is more than four times higher, backed by revenue growth that roughly matches the multiple. This isn't a story of a valuation running ahead of the business — the two moved together.

Why inference, and why now

The pitch behind Fireworks and companies like it is straightforward: most enterprises don't need the single most capable model available for every task, they need a model that's good enough for a specific job, running fast and cheap at high volume. That's a different optimization problem than the one frontier labs solve, and it's created room for a specialized infrastructure layer to sit between foundation models and the companies deploying them.

Fireworks' own framing — "specialized intelligence" — leans into that: smaller, task-tuned, or fine-tuned models served efficiently, rather than routing every request through the most expensive general-purpose model on the market. As more companies move from experimenting with AI to running it in production at real volume, the economics of inference — latency, cost per token, uptime — start to matter as much as raw model capability.

Nvidia on both sides of the table

One detail worth noting: Nvidia is both a strategic partner and an investor here, having also participated in the round. Fireworks says the new capital will go toward expanding compute infrastructure, growing its engineering team, and deepening partnerships with cloud providers including Microsoft and Nvidia. For Nvidia, backing the infrastructure companies that drive inference demand is a direct hedge on its own core business — more efficient inference platforms still need chips, just used more efficiently.

The bigger picture

Funding headlines in AI have mostly gone to frontier model labs. This round is a reminder that a comparably large business has been building beneath them — one that doesn't need to win the capability race to be worth tens of billions of dollars. As the market matures past "which model is smartest" toward "which model runs this specific workload most efficiently," the infrastructure layer that Fireworks occupies looks less like a bet on any single lab's success and more like a bet on AI deployment continuing to grow regardless of who's winning the frontier race.