Fireworks AI raises $1.5 billion Series D at $17.5 billion valuation

Fireworks AI announced a $1.505 billion Series D funding round on July 16 at a $17.5 billion valuation, one of the largest AI infrastructure raises of the year, as the inference platform reports it now serves more than 40 trillion tokens per day.

What's new

The round was led by Atreides Management, Index Ventures, and TCV, with participation from Evantic Capital, Lightspeed Venture Partners, NVIDIA, 20VC, Bessemer Venture Partners, and Menlo Ventures, among others. Fireworks confirmed the deal was a "$1.505 billion Series D at a $17.5 billion valuation, led by Atreides Management, Index Ventures, and TCV."

Alongside the raise, Fireworks disclosed it has surpassed $1 billion in annualized revenue run rate — up roughly 5x year over year from its prior funding round — while daily token volume served on its platform nearly tripled over the same period, from about 15 trillion to more than 40 trillion tokens per day. The company said more than 95% of those tokens now come from models specialized on customers' own proprietary data rather than off-the-shelf general-purpose models, and named Cursor and Harvey among customers building on its platform. Fireworks said the new capital will go toward expanding compute infrastructure and growing its engineering team.

Context

Fireworks operates in the increasingly crowded inference-platform market alongside Together AI, Groq, and SambaNova, all of which have raised large rounds over the past year to build out GPU capacity and compete on serving cost and speed for both open- and closed-weight models. Together AI closed an $800 million round at an $8.3 billion valuation earlier this month, and Groq raised $650 million in June — putting Fireworks' $17.5 billion valuation well ahead of both on a per-dollar-raised basis, and signaling investors see a widening gap between platforms that simply serve tokens and ones, like Fireworks, that position themselves around building customer-specific specialized intelligence on top of open models.

Why it matters

The scale of the numbers Fireworks is disclosing — a $1 billion-plus revenue run rate and 40 trillion tokens served daily, up nearly 3x in roughly a year — is a concrete data point on how fast enterprise demand for customized, self-hosted AI inference is growing outside the four frontier chat labs. Fireworks' emphasis on proprietary, fine-tuned models accounting for 95% of its token volume also reflects a broader shift already visible elsewhere in the market: enterprises increasingly want to own and specialize the intelligence running their products rather than renting general-purpose model access outright, a trend that benefits infrastructure platforms positioned to help them fine-tune and serve custom models at scale rather than the base-model labs themselves.