Fireworks AI reportedly raising new round at $15B valuation, riding inference token boom
The AMW Read
Novelty is incremental (another mega-round for an existing player already in the corpus), but significance is cross-segment structural: the valuation validates inference infrastructure as a decacorn category and updates the capital-flow pattern across AI segments.
Fireworks AI reportedly raising new round at $15B valuation, riding inference token boom
Chinese-born, US-based inference platform Fireworks AI is in talks to raise a new funding round at a valuation of approximately $15 billion (roughly RMB 100 billion), according to Bloomberg. The four-year-old company, founded by Lin Qiao (乔琳), a former Meta engineering lead who co-created PyTorch and Caffe2, provides a high-performance cloud platform for running open-source AI models on a per-token basis. Fireworks AI now processes about 30 trillion tokens daily and counts Cursor, Uber, Samsung, Notion, and Shopify among its customers. Its annualized revenue reached $315 million as of February 2026, up 416% year-over-year. The company has previously raised from Benchmark, Sequoia Capital, NVIDIA, and AMD, and earlier this year acquired game-server hosting firm Hathora to bolster its distributed-compute orchestration capabilities.
This round, if confirmed, would place Fireworks AI among the highest-valued AI infrastructure startups globally, alongside the mega-valuations seen at Anthropic ($965B post-H series) and OpenAI (targeting a $1T+ IPO). The reported raise exemplifies the "fastest-ARR-ramp" pattern and the infrastructure-layer value capture driven by the surging demand for inference compute, as enterprises and developers shift from training to deploying AI at scale. Fireworks AI occupies the inference-cloud niche, competing with hyperscaler clouds and other inference providers; its growth signals that the market for token-based AI infrastructure is expanding rapidly enough to support decacorn-level valuations.
The raise updates the inference infrastructure player map and validates the hypothesis that a specialized, open-model-optimized inference layer can achieve hyperscale economics without being a foundation-model lab. NVIDIA CEO Jensen Huang has previously called Fireworks AI "the TSMC of AI" — a comparison that underscores its role as a neutral, high-throughput compute fabric for multiple model ecosystems. However, the $15B valuation against $315M ARR implies a revenue multiple near 48x, which places this squarely inside the ongoing debate about whether current AI valuations represent a structural transformation or a capital-driven bubble. The company’s ability to sustain this multiple will depend on whether token growth continues to outpace revenue dilution and whether it can defend its margin against hyperscaler commoditization.
