By provider
| Model | Routing / Hosting | Calls | Tokens in | Tokens out | API-equiv $ |
|---|---|---|---|---|---|
| claude-opus-4-7 | US (Anthropic) | 771 | 4,026,893 | 782,027 | $511.9405 |
| openrouter/z-ai/glm-5.1 | US-routed (OpenRouter), CN-hosted (Z.ai) | 48 | 8,187,978 | 877,283 | $6.4334 |
| gpt-5.5 | Unknown | 1 | 67,923 | 787 | $0.0270 |
VPS bandwidth (rough proxy for non-API traffic)
| Direction | GB | Note |
|---|---|---|
| VPS RX (incoming) | 6.45 | OpenRouter responses, GitHub clones, npm, Granola transcripts, Postgres replies, etc. |
| VPS TX (outgoing) | 0.64 | OpenRouter prompts, dashboard serves, Postgres queries, Neotoma writes, etc. |
The policy push
This is not a fundraiser. It is a critical-infrastructure problem the way water and power are critical-infrastructure problems. Token spend on US-routed inference is the only thing keeping our stack defended right now, until baremetal sovereign infra lands. That transition will take a decade or more. The waitlists for the hardware that closes the gap are 18-24 months. The decision window is now.
Federal asks
- National AI compute strategy with explicit Blackwell / MI300X allocation, paired with the existing Pan-Canadian AI Strategy.
- Treasury Board sovereign-cloud framework: list approved Canadian-hosted inference providers (Cohere, Canadian DC operators) and require federal departments to default to them where feasible.
- Innovation, Science and Economic Development Canada (ISED) to coordinate inter-departmental GPU procurement at federal scale rather than per-ministry.
Provincial asks (Alberta as the obvious lead)
- Province-led data-centre siting with hydro / flare-gas / nuclear power agreements. Alberta's existing data-centre cluster around Calgary is a natural foundation.
- Provincial-scale GPU pre-orders. The federal procurement window is large; provincial windows can move faster.
- Alberta + AB AI Institute (AMII) joint sovereign-inference benchmark suite. Test Cohere, Canadian-hosted variants, and US providers head-to-head on accountability workloads like the ten Agency 2026 challenges.
Local asks
- Municipal co-location offers: Calgary, Edmonton, Toronto, Montreal, Halifax all have power + cooling capacity if zoning unlocks. Tax credits for sovereign-DC anchor tenants.
- Municipal data sovereignty by-laws: any municipal vendor running AI on resident data must disclose hosting location and routing.
Why now (the Mythos-class context)
Frontier agentic models are at the point where attack surface scales with capability. Token-spend rate-limiting and monitored API endpoints are the temporary defence. The durable defence is baremetal infrastructure with controlled attack surface — sovereign datacentres, Canadian-hosted models, audited supply chain. That posture takes a decade to stand up and the GPU waitlists are the gating step. Get on them now.