Vancouver, British Columbia
Frontier NVIDIA B300 compute in Vancouver, built for Canadian companies. No hyperscaler in between.
The constraint
Across Canada, AI infrastructure is being announced in gigawatts and delivered in years. New campuses, new substations, new timelines. Most of it is a bet that if the capacity gets built, the customers will arrive.
British Columbia does not work that way. BC Hydro has capped new power for AI data centres at 300 megawatts over two years under a competitive allocation, against a base of roughly 60 megawatts serving data centres across the entire province today. Allocation decisions land in September 2026. Until then, and for most applicants after, the answer is no.
That changes what matters. In a market where anyone can get power, the constraint is finding customers. In a market where power is rationed, the constraint is being already connected.
Freyr's capacity is contracted, powered and operating in Vancouver. Not a permit, not a phase two, not a 2029 delivery date.
The product
Freyr Technology Canada operates NVIDIA B300 nodes in Vancouver data centres. Eight GPUs per server, 288GB of HBM3e per GPU, 2.3 terabytes of GPU memory per node, connected by fifth-generation NVLink at 1.8 TB/s and 800G ConnectX-8 networking.
You get the machine. Bare metal, no hypervisor between your code and the silicon, no other tenant on your hardware, no orchestration layer deciding what you are allowed to run. We configure the cluster to your architecture, your storage profile and your interconnect topology before it goes into the rack.
That is the whole product. We are not selling you a platform, a model catalogue or an abstraction. If your team knows what it needs, we build it and hand you the keys.
288GB of HBM3e per GPU, 2.3TB per node. The highest-VRAM NVIDIA Blackwell available for rent in British Columbia. Models that will not fit elsewhere fit here without quantisation.
Your nodes are allocated to you for the length of the contract. Not pooled, not oversubscribed, not reallocated when a larger customer arrives.
Bare metal. No hypervisor overhead, no noisy neighbours, no throttling. The cluster performs the same in year three as on day one.
Configured to your specification and deployed in five weeks. No waitlist, no queue position, no capacity request form.
The structure
Every major provider sells flexibility as the virtue. Spin up, spin down, pay for what you use. For a workload that runs occasionally, that is the right answer.
For a workload that runs continuously, it is a different deal than it looks. Your instances exist at the provider's discretion. Capacity you had last quarter may not be there next quarter. Performance moves with whoever else is on the box. Price moves with the market, and the B300 market has moved 57 percent in nine months.
Freyr does the opposite. You contract the hardware for a defined term, one to five years. The rate is fixed for that term. The machine is not shared and cannot be reassigned. You know what you have, what it costs and what it will do, for the length of your roadmap.
That is a larger commitment than clicking a button. It is also the only structure that makes frontier compute a plannable line item rather than a variable one.
Where it runs
Freyr's compute runs in eStruxture's VAN-2 and VAN-3 facilities, Canada's largest domestic colocation platform, with high-density power and cooling built for AI workloads. Network presence at Cologix in downtown Vancouver puts us in the Harbour Centre carrier hotel with direct access to the Vancouver Internet Exchange and onramps to AWS, Azure and Google Cloud.
British Columbia's grid is hydroelectric, so the power behind your cluster is renewable without a generation project attached to it. Latency across the Greater Vancouver region runs around 5 milliseconds. For teams with existing infrastructure in a hyperscaler region, the cloud onramps mean a hybrid architecture does not have to cross a border to work.
Data processed on Freyr hardware stays in Canadian facilities.
Direct allocation of frontier Blackwell silicon, not resold capacity.
Hardware under contract and colocation under agreement in Vancouver facilities operating today.
A Vancouver team, a Vancouver address, and a named engineer on every deployment.
Global footprint
Freyr builds and operates AI infrastructure across [regions].
Awaiting content: named projects, scale per site, and the commitments line.
Nodes are configured to customer specification before deployment, so allocation follows commitment rather than the other way around. The 2026 deployment window is open now.
Talk to
John Kinash
Business Development, Freyr Technology Canada
Tell us your model, your scale and your timeline.
We will tell you honestly whether we are the right fit.
