SPUR Compute brings GLM-5.2 online - among the first in Canada to serve Z.ai's new frontier open model
Spur Innovation Centre today turned on inference for GLM-5.2, the latest frontier open-weight model from Z.ai (formerly Zhipu AI), on its SPUR Compute platform. Z.ai released GLM-5.2 on June 13, 2026; SPUR brought it online from Canadian soil within days of the open-weight release - placing SPUR among the first operators in Canada to serve the model.
GLM-5.2 is a 753-billion-parameter Mixture-of-Experts model (about 40 billion parameters active per token) with a stable 1-million-token context window and native tool-calling. It runs on 8 NVIDIA Blackwell B300 SXM6 GPUs hosted in SPUR's Waterloo, Ontario datacentre.
"Canadian organizations should not have to choose between frontier AI capability and keeping their data in Canada. GLM-5.2 is a top-tier open model - and within days of its release we have it serving from Canadian-owned hardware, under Canadian law, at a fraction of closed-lab token prices. That is exactly the sovereign-compute thesis SPUR was built on."
- SPUR Compute team
The technical fact-sheet
- Model
- GLM-5.2 (Z.ai) - 753B total parameters / Mixture-of-Experts (~40B active per token), 1M-token context window, native tool-calling, MIT-licensed open weights. Released 2026-06-13.
- Hardware
- 8x NVIDIA B300 SXM6 (Blackwell), HBM3e, served FP8 with tensor-parallel sharding across all 8 GPUs via vLLM.
- Hosting
- SPUR Innovation Centre datacentre, Waterloo, Ontario. Sovereign Canadian hosting, PIPEDA-aligned, no US CLOUD-Act exposure.
- Pricing
- Flat, transparent per-token pricing - see ai.spuric.com/models for live rates. Early independent testing has reported GLM-5.2 matching or beating leading closed models on long-horizon coding at roughly one-sixth the cost.
- API
- OpenAI-compatible REST API at https://ai.spuric.com/v1 - a drop-in replacement for any existing OpenAI SDK integration.
- Sign-up
- https://ai.spuric.com/signup - about 60 seconds from email to first API key. Free tier with starter credits.
Why this matters
GLM-5.2, released by Z.ai under the permissive MIT license, is among the strongest open-weight models available in 2026 - with a usable 1-million-token context window (five times its predecessor) and a coding-first design. Its "IndexShare" architecture reuses a single indexer across every four sparse-attention layers, cutting per-token compute substantially at the maximum context length.
Until now, Canadian organizations wanting frontier-grade inference faced a hard tradeoff: use a foreign-hosted closed-lab API (best capability, but data leaves the country and falls under foreign law), or self-host a smaller open model (data stays home, but a real capability gap). SPUR Compute closes that gap by hosting frontier open-weight models on Canadian-owned hardware in a Canadian datacentre. Prompts and outputs never leave the country.
The hardware story
SPUR serves GLM-5.2 on NVIDIA's B300 SXM6 (the second wave of the Blackwell architecture) using vLLM with FP8 weights and KV cache and full tool-calling enabled. SPUR's B300 nodes were among the first Canadian deployments of the generation, brought online in early 2026. The same fleet previously hosted DeepSeek V4; GLM-5.2 is the latest model SPUR has stood up on it, typically within days of a major open-weight release.
Sovereignty and compliance
- PIPEDA-aligned by construction. SPUR Compute runs entirely on Canadian-owned infrastructure inside SPUR's Waterloo datacentre. No third-party hyperscaler, no cross-border data flow.
- No US CLOUD Act exposure. SPUR is a Canadian-incorporated entity with no US subsidiary structure. Data and metadata cannot be lawfully subpoenaed by US authorities.
- No telemetry, no training. Customer prompts and outputs are not logged beyond the per-request counters required for billing, and are not used to train future models. Both are written into the customer terms, not opt-in toggles.
- Real-iron transparency. Every host machine has a named datacentre location and is uptime-verifiable on the public marketplace at ai.spuric.com.
About SPUR Innovation Centre
SPUR Innovation Centre is a venture studio and sovereign Canadian compute provider based in Waterloo, Ontario. The Centre operates a multi-rack GPU datacentre featuring NVIDIA Blackwell B300 SXM6, RTX 5090, and RTX 4090 hardware, and provides on-demand and reserved GPU compute, inference APIs, and full-stack venture services to early-stage Canadian companies.
SPUR was founded by Kapley Judge and operates a 700,000-square-foot campus in Waterloo (the former BlackBerry HQ), housing both compute infrastructure and venture-studio tenant companies.
Resources for media
Press contact
SPUR Compute Communications
Get in touch through our live chat: ai.spuric.com/chat - tell us you are press or a partner and we will route you.
SPUR Innovation Centre, 2240 University Ave, Waterloo, ON N2K 0A9
Available for technical interviews on the engineering stack, the business model, or the sovereign-compute thesis.
- ends -