Key Partners
- NVIDIA — investor + infrastructure partner narrative (FACT Series D / press).
- Open-weight labs — GLM, Kimi, DeepSeek, MiniMax, Qwen, Llama, Nemotron (FACT model library / homepage).
- Series D leads — Atreides, Index, TCV; Nvidia, Lightspeed, Evantic, Bessemer, Menlo, others (FACT blog).
- Cloud surfaces — Azure Foundry customer stories (FACT testimonials).
- Named customers — Cursor, Harvey, Notion, Vercel, Quora, Sourcegraph, Cresta, UiPath (FACT site/blog).
- OpenAI- and Anthropic-compatible API ecosystem for tool/SDK partners (FACT docs).
Key Activities
- Operate serverless multi-model inference (Standard / Priority / Fast).
- Provision On-Demand dedicated deployments (multi-region, custom models).
- Ship Managed Training, Serverless Training API, Dedicated Training API.
- Advance disaggregated serving stack (company inference narrative).
- Publish pricing + docs; Batch, cache, Reserved Throughput.
- Sell Reserved Capacity / enterprise quotas and BYOC.
Key Resources
- GPU SKUs — H100, H200, B200, B300, GB300 (FACT pricing tables).
- Disaggregated inference engine narrative (FACT inference page).
- 250+ models; day-zero launch partner posture (FACT inference FAQ).
- Train→serve loop — fine-tunes serve on same kernels (FACT inference).
- SOC 2 Type II / HIPAA-ready / GDPR claims (FACT FAQ).
- Named production customer relationships (FACT site/blog).
Value Propositions
- Serverless — popular open models, no cold starts, pay per token (FACT docs).
- OpenAI- and Anthropic-compatible APIs — migrate by swapping a URL (FACT).
- Cached input ~50%; Batch −50% on Standard rates (FACT docs).
- Priority + Fast serverless modes for reliability/speed (FACT docs).
- On-Demand dedicated GPUs for custom/LoRA and predictable throughput (FACT).
- Train and serve on one stack (SFT/DPO/RFT + Serverless Training API) (FACT).
- Company claim — specialized models dominate token mix; cost narrative vs closed APIs (blog).
Customer Relationships
- Self-serve signup + $1 free credits + public pricing (FACT site).
- Docs-led expansion — serverless → on-demand for LoRA/custom (FACT docs).
- Batch/cache education to keep cost competitive (FACT docs).
- Priority/Fast as reliability and speed upsells (FACT docs).
- Reserved Capacity / Throughput + Contact sales (FACT inference).
- Reference customers named on site and Series D post (FACT claim).
Channels
- Primary — fireworks.ai + /pricing public cards.
- docs.fireworks.ai quickstarts and serverless pricing tables.
- OpenAI- and Anthropic-compatible API / SDK discovery.
- Company blog — funding and specialized-intelligence narrative.
- NVIDIA + Azure Foundry partner/customer surfaces.
- Contact sales for Reserved Capacity, BYOC, enterprise quotas.
Customer Segments
- App/agent builders on serverless open models (FACT docs ICP).
- AI-native production companies (named site logos — FACT claim).
- Teams needing Priority/Fast or dedicated/custom endpoints (FACT).
- ML orgs training specialized models (SFT/DPO/RFT) (FACT pricing).
- Residency/enterprise buyers (US serverless, reserved, BYOC) (FACT).
- Coding/legal/agent products attaching train→serve (FACT testimonials).
Cost Structure
- GPU / power / colo COGS behind serverless + on-demand (INFERENCE).
- Peak-capacity cost behind Priority/Fast and US premiums (INFERENCE).
- Serving-stack and kernel R&D (company inference narrative).
- Training compute COGS for SFT/DPO/RFT (INFERENCE from pricing).
- Series D-funded capacity + eng expansion (FACT blog claim).
- GTM — PLG content + enterprise sales (INFERENCE).
- No public Fireworks AI P&L line invented.
Revenue Streams
- Serverless Standard — PAYG tokens; $1 free credits (FACT pricing).
- Examples — GLM 5.2 $1.40/$0.14/$4.40; Kimi K2.7 Code $0.95/$0.19/$4.00; MiniMax M3 $0.30/$0.06/$1.20 (FACT docs).
- Cached input discounts; Batch 50% of Standard (FACT docs).
- Priority / Fast premium token cards (FACT docs).
- On-Demand — H100/H200 $8/hr; B200 $13; B300 $15; GB300 $20 (FACT).
- Training — LoRA/full SFT & DPO $/1M bands; RFT = GPU-hr; Serverless Training API (FACT).
- Reserved / Enterprise — Contact sales for guaranteed capacity (FACT).