Fireworks AI Raises $1.5B Series D for Enterprise Custom AI

Fireworks AI has secured $1.5 billion in Series D funding, rocketing the company to a reported $17.5 billion valuation. The round was co-led by Atreides Management, Index Ventures, and TCV, with notable strategic investments from NVIDIA and TIME Ventures, as detailed in the company's Series D blog post. This investment underscores the intense market focus on infrastructure that enables large enterprises to customize open-weight AI models with proprietary data while maintaining predictable costs.

Fireworks Raises $1.5B Series D: fresh capital, fresh targets

Fireworks operates what industry analysts at Sacra call a B2B managed inference cloud. The platform is built to process tokens at massive scale for a growing number of customers, including industry leaders like Uber and Shopify. Its usage-based pricing covers the entire model lifecycle, allowing enterprises to pay per token for inference and fine-tuning, or per GPU-hour for reinforcement learning. This efficiency is driven by proprietary kernels like FireAttention and FireOptimizer, which can significantly boost throughput on many models. This technological edge allows Fireworks to offer competitive pricing against self-hosting while charging a premium. Significantly, a substantial majority of tokens are reportedly used for models already customized with client data, indicating a strong, recurring revenue stream.

Fireworks AI provides a cloud platform for businesses to customize open-source AI models using their own private data. This enables companies to create specialized, cost-effective AI applications while maintaining control over their intelligence, offering a managed solution for fine-tuning, training, and deploying models at scale.

Enterprise appetite for customization

Positioning itself as an "AI cloud for developer teams," Fireworks empowers organizations to own their AI capabilities rather than just renting them. The platform's key features are designed to meet this demand:
- Fine-tuning models on proprietary text and code
- Creating reinforcement learning loops from user feedback
- Providing dedicated deployments for sensitive data workloads
- Integrating with Microsoft Foundry to streamline procurement through Azure

The recent launch of Serverless 2.0 in June 2026 further addresses enterprise needs by offering enhanced reliability and speed controls without requiring long-term reservations. This move caters to the growing trend of large organizations using a multi-model strategy, where teams balance cost and performance by running three or more model families concurrently.

Growing share of enterprise budgets

The company's financial growth reflects its increasing importance to enterprise clients. According to Sacra, Sierra hit $200M in ARR in May 2026, up from ~$130M at the end of 2025 (and $26M at the end of 2024). Fireworks also reports that its token processing volume now surpasses that of some major proprietary model providers. Investors note a strategic shift in revenue, with high-value dedicated deployments and voice agent tools expanding the company's reach into new budget areas like contact centers and field operations. To fuel this growth, the company plans to significantly expand its engineering team, focusing on multimodal inputs, low-latency voice AI, and agentic task fine-tuning. The equity investment from NVIDIA may also signal deeper collaboration on future hardware like Blackwell GPUs, potentially lowering inference costs further.

Competitive landscape and risk factors

While competitors like Together AI and Baseten offer similar inference cloud services, Fireworks' reported token scale and revenue growth currently place it ahead. However, industry experts caution that the market remains dynamic. Enterprise customers continue to diversify their AI portfolio, utilizing models from Anthropic, Google, and open-source Llama variants. This trend suggests that for Fireworks to sustain its momentum, it must protect its position as a neutral, open-model platform and maintain its cost and performance advantages.