Fireworks AI Raises $1.5B at $17.5B Valuation as Enterprises Chase Cheaper, Specialized Models

Fireworks AI, the inference platform founded by former Meta PyTorch lead Lin Qiao, has closed a $1.5 billion Series D at a $17.5 billion valuation — one of the largest infrastructure rounds of the year and a marker of how quickly value is accruing to the layer that serves models rather than the labs that train them.

The round

Atreides Management, Index Ventures and TCV led the financing, with participation from Nvidia, Evantic and Lightspeed. The raise lands as Fireworks crosses $1 billion in annualized revenue run rate, up fivefold year over year — the kind of growth that explains a valuation that has itself multiplied several times since its last round.

Fireworks provides cloud-based serving for open and customized models, processing roughly 40 trillion tokens daily for customers including Elastic, GitLab and MongoDB.

The specialization thesis

The most telling statistic in the announcement: approximately 95 percent of tokens processed on Fireworks now come from customized models rather than off-the-shelf ones. Enterprises are taking open bases — Llama, Qwen, DeepSeek, Kimi — and tuning them into specialized systems for individual business processes, then paying Fireworks to serve them fast and cheap.

That positions the company as a direct beneficiary of this week's open-weight wave. Every major release — DeepSeek V4's stable version on July 24, Kimi K3's weights on July 27 — expands the catalog of frontier-class raw material Fireworks can serve the moment it drops, without paying a training bill. The Chinese open-model surge, in other words, is Fireworks' supply chain.

The infrastructure land grab

The round confirms the pattern running through 2026 venture data: with 70 percent of Q2 capital going to AI, investors are increasingly funding the plumbing rather than the models, anticipating that model capability commoditizes while serving, routing and customization compound. Fireworks now sits alongside Together AI ($800 million at $8.3 billion) and a crowded field of inference challengers — all betting that in the Flash-tier price war reshaping the industry, the arms dealer beats the armies.