Nvidia-backed Fireworks AI said Thursday it has raised $1.5 billion at a $17.5 billion valuation after annualized revenue surged past $1 billion, a fivefold increase from a year earlier. The financing underscores a shift in enterprise spending: finance chiefs are pressing teams to evaluate open-weight models hosted on independent inference clouds rather than defaulting to the closed systems sold by OpenAI and Anthropic, which investors have valued above $800 billion each.

Fireworks operates in the inference layer, managing the infrastructure that serves models to applications, and competes directly with Amazon Web Services, Google Cloud and Microsoft Azure as well as startups such as Baseten and Together AI. The company has also begun supplying graphics processors for model training, putting it in the same neocloud category as CoreWeave, which raised $1.5 billion in an initial public offering last year and now carries a $42 billion valuation, and Lambda and Nebius. DigitalOcean, another infrastructure vendor courting AI workloads, has seen its shares climb 149 percent this year as growth accelerated.

Chief executive Lin Qiao, a former Meta director who founded Fireworks in 2022 with six colleagues, described demand as super-linear in an interview at the company’s San Mateo headquarters. The platform lets developers plug in models from Chinese labs such as DeepSeek, MiniMax and Z.ai alongside open-weight releases from OpenAI, then fine-tune them with proprietary data. Qiao said the approach delivers specialized intelligence at five to ten times lower cost than equivalent closed models, a claim echoed by Coinbase chief executive Brian Armstrong in a June post on X.

The cost argument is gaining traction at the highest levels. Microsoft chief executive Satya Nadella wrote in a Sunday blog post that a company should be able to use a model without surrendering the knowledge that makes it unique, referencing remarks Palantir chief executive Alex Karp made on CNBC earlier this month. Karp said technical customers want to own the means of production rather than transfer it to a frontier lab.

Fireworks is not ceding distribution to the hyperscalers. In March it announced a partnership with Microsoft that allows Azure customers to access models through Fireworks, which draws compute from more than 20 suppliers including Microsoft itself. Qiao said the arrangement gives the startup much bigger reach without building its own data centers.

The company employs roughly 200 people and expects headcount to reach 600 by the end of 2026. Former Salesforce executive George Hu joined as president in April to build a formal sales organization after years of self-serve adoption. Fresh capital will also fund additional GPU capacity as Fireworks races to meet demand that Qiao says is only beginning to hit the gas.