This week the build-out went international while the rules tightened at home. Meta broke ground on a 1 gigawatt campus north of Edmonton, its first data center in Canada and a roughly US$9 billion bet on North American AI capacity. At the same time, Virginia switched on the first tax in the country aimed squarely at data center power use, several US communities said no to new projects, and utilities warned that the equipment to feed all of this, starting with transformers, is getting harder to get. Capacity is expanding across the map, and the constraints on it, power, permits, and parts, are moving to the front of every conversation.

For vendors and service providers, the signal is consistent with what we have flagged all quarter. The projects that move are the ones with a credible answer on power and community approval, and the partners who get in early are the ones who can speak to lead times, behind-the-meter power, and delivery dates they can actually hit. If a project or account below is on your target list and you want the people and timeline behind it, reach out and we will pull the thread.

For vendors, a gigawatt-scale campus of this size resets the demand picture on both sides of the border. It pulls power equipment, cooling, and fit-out scope into a single Canadian market that has not seen a build at this scale before, and it signals that the hyperscalers are now willing to plant gigawatt anchors well beyond the established US hubs. Partners who can serve cross-border logistics and long lead-time gear are positioned to benefit.

Microfluidic cooling moves the coolant onto the chip itself. As AI silicon pushes past 1,000 watts per chip and rack densities climb toward and beyond 130 kW, the industry is looking past cold plates to microfluidics, etching tiny channels directly into the back of the silicon so coolant flows across the heat source rather than through insulating layers. In testing, Microsoft, working with the Swiss startup Corintis, reported cooling performance up to three times better than advanced cold plates and peak temperature reductions of up to 65 percent, using AI-designed channel patterns modeled on the branching veins of leaves. The approach could raise server density, cut the energy spent chilling coolant, and eventually enable 3D-stacked chips that would otherwise overheat.

Source: Tom's Hardware and IEEE Spectrum, 2026.

A look at notable North American gatherings on the horizon. Windows are general; please confirm exact dates on the DCR calendar before making travel plans.

Planning your event schedule? See what DCR is attending and where we can connect.

Please confirm all dates on the DCR calendar, as conference schedules can shift.