Data centres for artificial intelligence draw enormous electric loads. For example, the data centre proposed by Telus in downtown Vancouver will draw about 100 Megawatts of power from the electric grid. All that power is eventually turned into heat, and in the Telus case this heat will be fed into a district heating system. Just How sustainable are Vancouver's AI data centres going to be? was explored in a recent Vancouver Sun article by Derrick Penner. Beyond the waste heat problem, rawing 100 MW of power still leaves important questions about how such large loads are managed when we reach peak demand hours in the winter. How can data centres provide demand flexibility? And should BC's provincial government institute mandated flexibility for new high-load connections, as other jurisdictions have already done? And if so, how much demand flexibility is needed? These are both conceptual and empirical questions that need answers going forward.

What exactly is mandated flexibility? Mandated flexibility takes the form of a flexible connection agreement as a precondition for approval of the connection. These agreements can contain a curtailment obligation where the data centre accepts a load reduction instruction from the utility during a predefined response-time, which can be as short as 15-30 minutes. Some agreements also distinguish between firm and non-firm capacity. Firm capacity must be provided, but non-firm capacity can be curtailed without compensation during a specified number of hours per year (typically 200-500). There are usually also provisions for safeguarding hardware when power is curtailed. Increasingly, mandated flexibility is brought in because voluntary flexibility can be insufficient. Flexible connection agreements include performance penalties for non-delivery, when firms do not respond to instructions by the utility as required. Such agreements can also narrow the response window by season and time of day, which gives data centres better planning ability. Utilities may also consider rewarding flexibility with a tariff structure that gives customers subject to non-firm capacity provision a discount on the non-firm capacity (but not the firm capacity).

The economic case for mandated flexibility rests on the opportunity cost of imposing constraints on other grid connections, or building new generation assets. For example, households are difficult to coordinate, and demand-side management (DSM) is rather challenging because there are few smart appliances and smart chargers that can be powered down during a critical time window. It is much easier to engage in a contract with large-load customers as they are single decision makers. Building new generation assets can take many years of planning and construction, and therefore this only provides a long-term solutino but not a solution in the short term.

Technically, there are solutions to institute demand response from lage-scale customers using appropriate software. For example, OpenADR 2.0 (Automated Demand Response) is a standardized protocol through which a utility can send instructions and signals directly to the data centre's energy management system in order to trigger responses without human intervention. A speedy response is often of high importance.

The enormous growth in data centres that is projected for the next years, in response to growth of artificial intelligence applications, will put significant new strain on electrical grids. (I had explored it in last week's blog about power demand in BC.) It is important for regulators to be prepared. Mandated flexibiity is one important way to ensure grid stability and reliability, especially during peak hours of electricity demand in BC during cold winter days. Data centres are particularly adept at providing flexibility as they can route demand to other data centres that do not face electricity supply constraints. Their ability to shuffle computing demand across multiple locations seamlessly enables managing demand flexibility for the electric grid.

The classic first-come-first-serve principle of connecting new customers to the electricity grid needs to be augmented to allow for a new large-load-with-flexibility principle. Data centres are not the only large-load customers who could provide demand response. Electric natural gas liquefaction trains, if they are built in BC with potential loads of 600-700 MW for a 12-13 million tonnes per year of LNG plant, ought to provide demand response as well. BC's provincial government should take note of what other jurisdictions have been doing.

In Ireland, data centres are estimated to draw 20-21 percent of the country's national electricity consumption, and that number is expect to grow beyond 30 percent in the next year. After imposing a moratorium on new connections in 2022, a new framework was brought in that requires data centres to commit to procuring 80% of electricity from nrewable sources, provide on-site dispatchable generation or storage, and locate where they don't add to grid congestion. These requirements are in lieu of explicit response flexibility. In the Netherlands, the Dutch tranmission system operator TenneT introdouced time-dependent transmission rights (TDTR) contracts, which offer customers the option to buy residual capacity outside peak hours as a "non-firm access proudct". In addition, new customers face penalties for peak-hour consumption. The European Union is amending Directive 2024/1711 to institute a binding obligation on member countries to develop a framework for system operators to offer flexible connection agreementrs. Germany's federal network agency ("Bundesnetzagentur") has been developing non-firm access products. The legal basis is section §14a of the Germany Energy Industry Act that empowers grid operators to temporarily reduce power consumption during peak periods.

In the United States, due to the fragmentation of the electric system into separate markets, different interconnections have different rules. Most notably the PJM interconnection, where many large data centres are clustered, ahs developed a non-capacity-backed load (NCBL) product that it being considered for adoption. And in Texas, with its ERCOT interconnection, Senate Bill 6 (SB6) was passed into law last year that mandates that new large loads exceeding 75 MW prove their load flexibility and agree to curtail power on demand during grid emergencies. SB 6 authorizes ERCOT to order data centres with backup generation to deploy those generators or curtail their power usage during grid emergencies.

Closer to B.C., Alberta moved last year to put new rules on connecting large new loads. Following the implementation of Bill 8 (Utilities Statutes Amendment Act), the Alberta Electric System Operator (AESO) introduced strict Large Load Integration programs to tie data center capacity limits to operational and emergency flexibility. Specifically, it implementes a "bring your own power" model that icentivizes data centres to supply their own power generators, requires data centre to cover the cost of necessary transmission upgrades, and that new grid connections are conditional on reliability and adequacy of supply. In addition, Bill 12 (Financial Statutes Amendment Act) introduces new levies on grid-connected data centres that exceed 75 MW capacity, with the levy set as a percentage (1-2%) of the computing equipment purchase cost.

Different jurisdictions pursue somewhat different solutions to the challenge of connecting new large loads. Even in British Columbia, where our hydro system provides significant supply flexibility, our supply is not limitless. Theare are hours each year where our system approaches peak capacity (see the data dtable in last week's blog). The province should start working with B.C. Hydro to chart a course that allows industries to grow, while protecting our electricity system from undesirable repercussion. We need empirical modelling to determine just how much demand flexibility is needed, as this may also involve local transmission congestion constraints in addition to supply constraints. My sense is that we need a new regulatory approach rather sooner than later.