Roadmap

We don't ship all five layers at once — each phase builds on the last: first prove the model business, then plug in compute supply, then sharpen the data and evaluation edge, then take it global, then layer on tools and orchestration.

Phase overview

Phase 1 — Domestic foundation (in progress)

Timing: current

Key work

• Launch the platform and get the domestic version running
• Aggregate domestic and overseas closed-source models into a full-category supply
• Build the sales channel network, validate the "model spread" revenue model
• Start relationships with compute centers to prep Phase 2

Goals

• Complete coverage of mainstream closed-source models; one platform, any model
• Close the loop: model call → compute consumption → billing
• Accumulate a first batch of loyal users and channel partners

Phase 2 — Open-source deployment + data evaluation

Timing: starts once Phase 1 revenue is validated

Key work

• Connect idle GPU capacity from compute centers and deploy popular open-source models
• Launch the "Where does compute come from" comparison: same open model, different data centers, real price / latency / concurrency
• Launch the "AI Barometer" call-volume rankings built from platform data
• Deliver the full closed + open matrix

Goals

• Become the domestic platform with the broadest model coverage
• Build industry credibility through comparison and rankings
• Help compute centers absorb idle capacity; validate the compute-absorption fee model

Phase 3 — Overseas replication

Timing: starts once closed + open + evaluation are proven domestically

Key work

• Launch the overseas site with the same scope as domestic
• Replicate the model aggregation + compute absorption + evaluation stack abroad
• Onboard overseas local closed/open models; ship Chinese-origin models (Qwen, DeepSeek, etc.) outbound through the platform
• Adapt to overseas payments, languages, and compliance

Goals

• Build the "first stop for Chinese models going abroad" brand
• Enable two-way global flow of models and compute
• Run the overseas site as a standalone P&L

Unique tailwind: OpenRouter data shows Chinese AI call volume has overtaken the US for several consecutive weeks. The overseas site is built to ride that demand.

Phase 4 — Tool & Agent ecosystem

Timing: starts once both sites have stable user bases

Key work

• Launch the tool marketplace / Agent store
• Onboard indie developers and small teams to list their agents and tools
• One-stop showcase, invocation, billing, settlement
• Build a standardized evaluation framework so users see real effectiveness

Goals

• Form the loop: developers ship → platform promotes → users call → developers get paid
• More tools bring more users; more users drive more model and compute consumption — classic network effect

Phase 5 — AI orchestration + compute operations

Timing: once the ecosystem is mature

Key work

• Launch the AI orchestration engine: users describe needs in natural language, AI picks the optimal plan
• Scale compute hosting and data-center operations as enterprise-grade services
• Build compliant AI data insights and industry reports on top of aggregated call-volume data

Goals

• Become the "intelligent operating system" of AI infrastructure
• Evolve from a light-asset matchmaker into a full-stack, hybrid-asset service platform