Opening: The Constraint Alberta Just Made Explicit
If you’re evaluating AI infrastructure in Alberta—or anywhere grid-constrained—the single most important shift to internalize is this: at today’s load levels, grid connection is not a checkbox. It is the gating factor. Capacity availability, interconnection timelines, and system impact requirements now shape capital deployment decisions more than land acquisition or facility design.
Alberta made that transition explicit in 2025. The Alberta Electric System Operator (AESO) imposed a reliability-driven cap limiting large load connections to just 1,200 MW between June 2025 and 2028, stating clearly that connecting all projects seeking access “would impair grid reliability.” The queue wasn’t slow—it was shut. By November 2025, that entire 1,200 MW interim allocation had been awarded to just two projects: 970 MW to the GLDC Load and 230 MW to the Keephills Data Centre Phase I. Everything else was deferred.
The backdrop makes the scarcity undeniable: 29 proposed projects representing more than 16 GW of demand were seeking connections—a scale the AESO stated Alberta had not previously experienced. When demand outstrips available capacity by more than 13×, you’re not managing a queue. You’re rationing infrastructure.
This is not unique to Alberta, but Alberta’s response is instructive because it couples the capacity cap with an explicit legislative and policy shift: Bill 8 (Utilities Statutes Amendment Act) and the province’s AI Data Centres Strategy now encourage developers to “bring their own generation” and apply cost-causation principles so that data centres—not ratepayers—pay for transmission upgrades. For anyone building 5–50 MW AI campuses, this is the governance architecture that now defines bankability, not the megawatt spec sheet.
Executive Signal: The Constraint Shift
What constraint is tightening?
Grid connection entitlement. Alberta’s AESO capped large load access at 1,200 MW through 2028 for reliability reasons, and awarded the full allocation to two projects. The remaining 16 GW queue is deferred indefinitely.
What operational shift is occurring?
Power is no longer a utility service you contract after site selection. It is the lead asset you must control—either through long-term power purchase agreements (PPAs), co-located generation partnerships, or behind-the-fence self-supply—before capital markets will underwrite the project.
Who must adjust?
Any organization modeling AI infrastructure as “digital real estate + grid hookup.” That financing logic is now structurally fragile in jurisdictions where connection rights are rationed and policy favors power-integrated solutions over load-only requests.
The constraint sequence is: Physics (grid reliability limits) → Capital (transmission upgrade costs) → Risk (curtailment and price volatility) → Governance (AESO caps and Bill 8 cost-causation framework).
What Seasoned Operators Are Seeing
Alberta’s transition from queue to entitlement is not theoretical. It is producing observable market behavior.
First observation: Connection rights are trading as assets.
In February 2026, Canadian media reported a “gold rush” for electricity allotments, with at least one allocation reportedly resold for substantial value. The commercial confirmation came from ChannelLife Canada’s reporting that Kalina Distributed Power sold “allocated megawatts” for CAD $18 million after transferring capacity originally allotted under the AESO’s interim process, with completion tied to execution of a Demand Transmission Service (DTS) agreement. When megawatts transfer for $18 million before a single rack is deployed, the market is signaling that connection entitlement itself has become monetizable infrastructure.
Second observation: The AESO queue is now a policy gate, not a service desk.
The AESO’s own public statements are unusually direct: connecting data centres and other large loads is “a rapidly evolving process” introducing “complex technical and operational challenges,” and loads without generation “significantly strain supply.” That language—on an official grid operator page—is not technical documentation. It is policy signaling. Alberta is telling you which projects will face friction.
Third observation: “Load-only” strategies are being filtered out.
The two projects that secured the full 1,200 MW allocation (GLDC at 970 MW and Keephills at 230 MW) were not generic colocation developments. They represent power-integrated or power-firmed models. The projects left in the deferred queue are the ones that could not demonstrate incremental supply or could not credibly navigate the AESO’s new prioritization logic.
For operators with institutional experience in regulated infrastructure, this pattern is familiar: when a public resource becomes scarce, governance shifts from service provisioning to allocation policy—and allocation policy always reflects broader political economy constraints, not just engineering capacity.
Technical Reality: Where Physics and Governance Intersect
The Reliability Cap and What It Actually Means
The AESO’s June 4, 2025 announcement is the authoritative source. It states the 1,200 MW interim limit is “the maximum additional large load capacity the grid can serve without negatively impacting grid reliability,” and exists specifically because connecting all loads seeking access would impair reliability. This is not a planning convenience. It is a hard constraint derived from system studies that model voltage stability, fault response, and generator dispatch adequacy under high-load scenarios.
The scale mismatch is extreme: 29 projects seeking more than 16 GW of capacity, against an interim allocation of 1,200 MW. That is a 13× demand oversubscription. For context, a competitive telecom spectrum auction might see 2–3× oversubscription. A 13× queue imbalance signals systemic misalignment between market expectations and physical grid capability.
Bill 8 and the BYOG Policy Architecture
Alberta’s legislative response is Bill 8 (Utilities Statutes Amendment Act), which the Government of Alberta describes as encouraging data centres to “bring their own generation” and aligning with cost-causation principles for transmission such that data centres—not ratepayers—pay for necessary upgrades.
This is governance translated into capital structure. The traditional model—where a utility socializes transmission expansion costs across the rate base and a developer pays only an interconnection fee—breaks down when speculative load reservations could trigger multi-hundred-million-dollar transmission builds with uncertain cost recovery. Bill 8 shifts that risk back onto the project proponent.
Bill 8 also ties into broader market redesign: it supports implementation of Alberta’s Restructured Energy Market, with implementation expected to begin in 2027 (temporary reliability and pricing measures were enacted July 1, 2024). The wholesale market structure matters because Alberta runs an energy-only market—generators are paid for energy produced, not separate capacity payments—and the AESO manages ancillary services for reliability. In an energy-only design, pool prices can be volatile, and large loads face direct exposure to supply-demand imbalances unless they hedge or self-supply.
The Utilities Consumer Advocate explains that Alberta’s competitive market sets pool prices where all dispatched suppliers receive the highest dispatched offer price—a design that can produce price spikes during tight supply conditions. The Government of Alberta enacted temporary Market Power Mitigation and Supply Cushion regulations effective July 1, 2024 to address reliability and offer behavior, underscoring that the market is under active policy intervention.
Transmission Cost Causation and the AUC’s Role
The Alberta Utilities Commission (AUC) regulates utilities and the siting of generation, transmission, and distribution facilities, and considers social, environmental, and economic implications for ratepayers when reviewing applications. The AUC explains that the ISO tariff (approved by the AUC) covers payments to transmission facility owners and other transmission-related costs, and applies province-wide to distribution companies and large industrial consumers—who then recover costs from customer classes through regulated rates.
Bill 8’s direction to move toward cost causation means that if your project triggers a new 240 kV line or substation expansion, you may be expected to fund it upfront or through a surcharge structure rather than having it amortized into the general rate base. This is not punitive—it is Alberta trying to prevent a political crisis where residential and commercial ratepayers subsidize speculative AI infrastructure that may or may not materialize.
For project finance, this changes everything. You can no longer model interconnection as a fixed, predictable fee. You must model it as a contingent capital obligation tied to transmission system impact studies—and those studies are now part of the AESO’s prioritization and allocation logic.
Capital & Risk Implications
Power Must Be the Lead Asset, Not the Utility
The clearest capital implication is that power strategy now precedes site selection and must be bankable before equity or debt will close.
In Alberta’s new framework, three power postures emerge:
Grid-only interconnect remains viable only for small-scale loads (5–10 MW) with schedule flexibility and business models that can tolerate pool-price volatility or that secure long-term hedges. For anything at 20–50 MW scale, treating grid-only as the default is a misread of Alberta’s policy direction.
Hybrid (grid + firming) becomes the rational middle path when you must reduce price volatility, protect uptime, and blunt queue and transmission-upgrade exposure while retaining a grid tie. The emergence of “power + land + load” deals—like TransAlta’s Keephills MOU with CPP Investment Board and Brookfield for phased data centre development backed by a long-term PPA for ~230 MW—signals that power firming is becoming the bankable structure.
Full BYOP (bring your own power) / self-supply becomes rational for 20–50 MW+ loads that are schedule-driven and willing to take on emissions and permitting obligations in exchange for power certainty. Synapse Data Center’s proposed 1 GW facility in Olds—claiming a self-supplied natural gas power system not connected to the provincial grid—illustrates how proponents are trying to align with Alberta’s stated “off-grid” preference while bypassing grid congestion entirely.
The Two-SPV Capital Stack Model
For projects at 10–50 MW scale, the financing structure that is emerging as standard is a two-SPV approach:
-
Power SPV: Generation assets, fuel arrangements, emissions compliance (TIER Regulation obligations), and long-term PPA or tolling contracts. This SPV underwrites against dispatchable capacity and contracted offtake, not pool-price speculation.
-
Facility/Colocation SPV: Land, shell, electrical/mechanical systems, cooling infrastructure, and interconnection. This SPV underwrites against lease or compute-as-a-service revenue, with power risk transferred to the Power SPV or hedged via contract.
Reference capital benchmarks (not project-specific, but directionally useful):
- Digital infrastructure (non-IT capex): JLL forecasts average global data centre construction costs rising to ~$11.3 million per MW in 2026, reflecting material cost growth from 2020–2025.
- Behind-the-fence gas combined cycle: Lazard presents a capital cost range of $650–$1,300/kW (~$0.65M–$1.3M per MW) as a reference before site-specific and decarbonization adders.
- Battery storage: Reuters reported average battery project costs around $125/kWh in late 2025, providing a basis for translating “hours of storage” into capex logic for ride-through and peak management.
The strategic shift is that power is no longer a line item in the facility budget—it is a parallel capital stack with its own SPV, its own counterparties, and its own risk profile. If your internal model still treats power as “a utility hookup,” it is structurally obsolete.
Stranded Queue Risk and Political Fragility
A simplistic strategy like “secure more MW in the queue as early as possible” is no longer just aggressive—it is politically and regulatorally fragile. Alberta is actively redesigning the rules to penalize exactly that behavior. Bill 8’s cost-causation framework and the AESO’s prioritization of projects with incremental supply mean that speculative queue positions without credible power plans will face increasing friction and potential disqualification.
The market has already priced this risk: the $18 million transfer of allocated megawatts was contingent on execution of a DTS agreement—meaning the buyer was paying for contractually secured entitlement, not a queue number. The implication is that queue positions without executed connection agreements are not bankable assets.
Execution Reality: Where Programs Break
Most infrastructure failures are coordination failures between physics, schedule, and governance. In Alberta’s AI data centre context, programs break at predictable points:
Underestimating substation and transmission dependencies.
A 30 MW AI campus does not just need 30 MW of generation capacity—it needs transmission infrastructure capable of delivering that load without voltage sag, harmonic distortion, or relay coordination failures. If the nearest substation is at 80% utilization and your connection triggers a substation expansion or new 240 kV line, you’ve just added 18–36 months and $50M–$150M to your critical path—costs that Bill 8 may now assign to you, not the rate base.
Misaligned rack density assumptions.
Designing for “aspirational” 20–30 kW/rack when your generation and cooling systems are scoped for 10–12 kW/rack creates stranded capacity risk. Conversely, under-provisioning thermal rejection or electrical distribution for higher-density AI workloads forces costly retrofits that can violate structural load limits in existing buildings.
Cooling retrofits violating structural limits.
Direct-to-chip liquid cooling and rear-door heat exchangers can materially reduce facility-level energy intensity, but they also introduce fluid management, leak detection, and maintenance complexity. Retrofitting legacy air-cooled facilities for liquid infrastructure often hits structural and spatial constraints that were not modeled in the original feasibility study.
Fabric oversubscription under AI traffic patterns.
Traditional enterprise data centre networks tolerate 3:1 or 4:1 oversubscription ratios on spine-leaf topologies. AI training and inference workloads generate east-west traffic patterns (model synchronization, distributed training across GPU clusters) that can saturate fabrics designed for north-south web traffic. Underestimating fabric bandwidth requirements leads to performance bottlenecks that undermine the entire compute investment.
Governance gaps between facilities and IT.
Energy strategy, emissions compliance, and transmission tariff exposure are typically managed by facilities or finance teams. AI compute scaling, model deployment cadence, and workload density are managed by IT or ML engineering teams. When those functions operate in silos, the result is misalignment between power procurement timelines and compute deployment schedules—leading to either stranded power contracts or delayed revenue ramps.
The common thread: Alberta’s new governance framework punishes projects that treat power as a utility service to be contracted after site selection. Disciplined execution now requires power strategy to lead, not follow, the site and design process.
What Disciplined Operators Are Doing Differently
Seasoned infrastructure operators are responding to Alberta’s governance shift by treating power as the anchor asset and site selection as a secondary optimization.
Designing by density class from day one.
Rather than specifying “a 20 MW facility,” disciplined operators are designing around power availability first: “We have access to 20 MW of firm, contracted power—what density class and cooling architecture optimizes capital efficiency and workload flexibility within that constraint?” This inverts the traditional design sequence but aligns to the new financing reality.
Modular 4–20 MW block planning.
Instead of single-phase 50+ MW developments that concentrate schedule and capital risk, operators are structuring phased deployments in 4–10 MW or 10–20 MW blocks. This matches the AESO’s allocation behavior (which favored a 230 MW initial phase at Keephills with potential to scale to 1 GW) and allows power procurement and generation buildout to sequence incrementally rather than committing full capex before revenue proof.
Energy treated as governance, not procurement.
Power strategy is now part of the executive governance layer—integrated with capital allocation, regulatory strategy, and stakeholder engagement—not delegated to procurement or operations. The Greenlight project illustrates this: securing 907 MW of AESO allocation, executing a DTS agreement, and targeting a first-half-2026 FID reflects a governance-first approach where power entitlement unlocks capital, not the reverse.
Fibre diversity engineered early.
While power is the binding constraint, disciplined operators are also front-loading fibre diversity analysis. AI campuses are increasingly serving as inference endpoints or training hubs, not just backend compute—which means network resilience (physically diverse fibre paths, multiple carriers, sub-10ms latency to major peering points) becomes a core site selection criterion alongside power.
Security segmentation aligned to uptime.
Operators are framing cybersecurity as availability architecture: macro-segmentation of AI workloads, identity boundaries aligned to compliance domains (ISO 27001, SOC 2, NIST frameworks), and failover/DR designs that treat security incidents as a fault domain rather than an afterthought. This is especially relevant in Alberta, where AESO reliability obligations and potential curtailment scenarios require clear isolation between critical loads and interruptible loads.
Board-Level Questions
These are the questions disciplined operators are forcing into executive and board discussions—not because they are comfortable, but because they are now financing gates:
-
What is our time-to-power versus time-to-procure gap?
If we need first power by Q2 2027 but the AESO queue is capped and Bill 8 cost-causation rules are still being finalized, what is our credible path to energization—and what is the contingency if that path fails? -
Are we designing for density class or aspirational specs?
Is our power and cooling budget based on contractually available capacity, or on a future-state assumption that may not survive AESO allocation or permitting delays? -
What is our campus-level fault-domain model?
If we lose grid supply, or if our behind-the-fence generation trips, what loads fail—and what is our contractual SLA exposure when that happens? -
Where is stranded capital exposure hiding?
Have we modeled the scenario where we secure land, complete civil and shell, but fail to close a DTS agreement or PPA—and if so, what is the carrying cost and exit strategy? -
Who owns the power SPV, and what is the counterparty risk?
If we contract with a generation partner via long-term PPA or tolling agreement, what happens if that partner defaults or if AESO curtails their dispatch—and how do we model that credit risk in our cost of capital? -
What is our emissions liability under TIER and federal OBPS?
If we self-supply with behind-the-fence gas generation, what is our annual compliance cost under Alberta’s Technology Innovation and Emissions Reduction (TIER) Regulation, and what happens if federal Clean Electricity Regulations tighten between now and commercial operation? -
Are we politically defensible if this becomes a rate-base fight?
If ratepayer groups or the AUC challenge our transmission upgrade allocation, can we demonstrate that we followed cost-causation principles and that our project does not socialize risk onto Alberta households and businesses?
Reynar Point of View
At Reynar IT, we advise clients that in Alberta, sustainable AI infrastructure begins with disciplined power strategy. This is not a slogan. It reflects the regulatory, financing, and grid environment that now shapes every serious data center investment in the province.
Alberta’s data center growth is real—and the opportunity is significant. But the defining advantage will not come from compute density alone. It will come from aligning land, power, and capital in a way that is financeable, resilient, and operationally scalable.
Operators who prioritize integrated power planning early—whether through grid coordination, behind-the-fence generation, or hybrid models—will move faster, secure capital more confidently, and deliver infrastructure that stands the test of policy and market cycles.
Those who sequence power correctly don’t just build capacity. They build certainty. And certainty is what makes infrastructure bankable.
About Reynar IT
Reynar IT provides executive infrastructure strategy across AI factories, modular data centers, and mission-critical enterprise platforms. We help operators and public-sector organizations align power, architecture, and governance to deliver scalable, long-term infrastructure with confidence. Learn more at r-it.ca.