Key Partners
- Cloudflare — joining / parent narrative Nov–Dec 2025 (FACT blogs).
- Catalog publishers — Black Forest Labs, Ideogram, Recraft, DeepSeek, Wan (FACT pricing).
- Community model creators contributing thousands of models (FACT homepage/pricing).
- Cog open-source packaging for custom models (FACT homepage).
- NVIDIA GPU SKUs in public hardware table (FACT pricing).
- Enterprise “trusted by” customer logos (FACT marketing claim).
Key Activities
- Serve public predictions on shared hardware pools.
- Host official + community model catalog with cost estimates.
- Run fine-tuning / training jobs (e.g. FLUX).
- Deploy custom Cog models; autoscale private traffic.
- Operate Deployments (min/max instances, stable endpoints).
- Publish pricing/docs; sell Enterprise volume & SLAs.
Key Resources
- GPU SKUs — T4, L40S, A100, H100, H200 (+ multi-GPU) (FACT pricing).
- 50,000+ model catalog claim (FACT enterprise).
- Cog open-source packaging + generated API servers (FACT homepage).
- Public vs private vs Deployment billing modes (FACT billing).
- Client libraries / HTTP API / docs (FACT docs).
- Cloudflare Developer Platform adjacency (FACT blogs).
Value Propositions
- One-line API / web Run for production-ready models (FACT homepage).
- Public — pay active time only; setup/idle free (FACT billing).
- Transparent hardware $/sec and per-output estimates (FACT pricing).
- Fine-tune without owning multi-GPU clusters (FACT docs).
- Cog custom models with autoscale (FACT homepage/pricing).
- Deployments for stable endpoints and cold-boot control (FACT billing).
- Enterprise — one contract, 50k+ models, SLAs (FACT enterprise).
Customer Relationships
- Self-serve signup + public pricing cards (FACT site).
- Model pages show run-time and cost estimates (FACT docs).
- Docs-led expansion — public → Deployments when steady (FACT billing).
- Prepaid credit or monthly arrears billing (FACT billing).
- Enterprise — dedicated AM, priority support, SLAs (FACT enterprise).
- Community model publishing as relationship flywheel (FACT homepage).
Channels
- Primary — replicate.com model explore + Run UI.
- Public /pricing hardware and output examples.
- docs.replicate.com quickstarts and billing topics.
- Client libraries / HTTP API / OpenAPI.
- Cog open-source + GitHub Actions publish guides.
- Cloudflare blog/dev platform adjacency; Enterprise form.
Customer Segments
- App/creative builders on public model APIs (FACT homepage ICP).
- Products metering images/video/tokens via official models (FACT pricing).
- Teams fine-tuning bases like FLUX (FACT docs).
- Engineers deploying custom Cog models (FACT homepage).
- Prod apps needing Deployments / no shared queue (FACT billing).
- Enterprises — one contract, SLAs, reserved compute (FACT enterprise).
Cost Structure
- GPU / power / colo COGS behind predictions (INFERENCE).
- Absorbed setup/idle on public shared pools (INFERENCE from billing).
- Dedicated capacity for private models & Deployments (INFERENCE).
- Platform/Cog/serving R&D (INFERENCE from product shape).
- Catalog and community operations (INFERENCE).
- GTM — PLG content + Enterprise sales (INFERENCE).
- No public Replicate P&L line invented.
Revenue Streams
- Public predictions — active GPU-sec or per-output; setup/idle free (FACT).
- Hardware examples — T4 $0.81/hr · A100 $5.04 · H100 $5.49 (FACT).
- Official meters — FLUX Pro $0.04/img; Wan 480p $0.09/s video (FACT).
- Fine-tune training — e.g. FLUX 8×H100 $0.0122/sec (FACT docs).
- Private models — setup + idle + active (fast-boot exception) (FACT).
- Deployments — bill online instances; autoscale (FACT billing).
- Enterprise — volume discounts, SLAs, reserved compute (FACT).