The interesting number in Alberta’s data-centre power story is not the largest announcement. It is the gap between what has been requested and what has been allocated.

As of 30 July 2026, the Government of Alberta cites about 19,565 MW of large-load requests. 2025 maximum demand on the Alberta system was about 12,800 MW. The request book is larger than last year’s peak. That is a fact about published figures. It is not a build programme.

Allocation is the scarce object

AESO Phase I large-load allocation is 1,200 MW, and it is fully allocated: 970 MW to GLDC for the Sturgeon / Meta campus, and 230 MW to Keephills. Those contracts are materially stronger than a connection request. Methodology treats an AESO request as not an allocation.

Even inside Phase I the two names are not the same project. Meta has a master site permit, site preparation, a public capital commitment, contracted Capital Power supply, and an approved dedicated Greenlight plant with FID. Keephills has the strongest non-construction power signal in the tracker - and TransAlta says no specific data centre is yet being built. A load contract can exist before a hall does.

The tracker is not a second queue total

The DC Intel tracker as of 11 September 2026 lists 33 projects and 17,665.8 MW of announced capacity across 31 rows with a published figure. Those MW mix facility power, computing load, on-site generation and connection requests. They should not be treated as committed capacity, and they should not be added to the 19,565 MW request figure as if both counted the same thing.

Alberta Data Centre Watch, cited in the same research workbook, publishes 13.1 GW of active data-centre demand from 20 of 30 active projects, and 8.0 GW of on-site generation from 13 active projects. Generation is separate from load. Do not add it to demand. Watch’s 13.1 GW and the tracker’s 17.7 GW announced are different published cuts. Neither is a forecast that the stack is built.

Behind-the-meter does not skip the gate

Alberta’s public grid materials describe bring-your-own-generation / Phase 2 pathways that prioritize dedicated new generation or storage. The process is still developing. Several Highly Credible and Probable names already rest on dedicated gas: Bitdeer’s approved on-site plant, Meta’s Greenlight plant, Beacon’s proposed Heartland and Indus generation, Teton’s behind-the-meter designs.

That is a strategic advantage only if the plant is permitted and financed. Synapse Olds is the negative control: a 1 GW campus concept whose associated power plant was denied by the AUC in August 2026. The proponent confirmed it would not proceed as proposed. Power can stop a project as well as enable it.

Goldfinch shows the other side of the same coin. 950 MW in the tracker is an AESO connection request (P3102), not a grid allocation. Land-use progress is real. The request is still a request.

What “likely to be built” can honestly mean

“Likely” here is not a percentage of the queue. The honest comparison is:

  • 1,200 MW Phase I allocated
  • about 19,565 MW requested as of 30 July 2026
  • five tracker projects under construction and one approved, on mixed MW bases that are much smaller than the announcement stack

Most requested load will not energize on the allocated-grid path that Phase I already closed. Some behind-the-meter projects may proceed without that path. This article does not forecast how many megawatts of those plants will reach commercial operation, or how much of the 24 proposed tracker names will reach FID.

For the scoring implication, see why power will decide the winners. For which names currently carry delivery evidence, see projects most likely to proceed.