May 13, 2026

AWS is telling customers to get out of its UAE cloud region—fast—after drone strikes damaged its data centres and knocked services offline for what could be months. That’s not a routine incident report; it’s a reminder that “region” is a geopolitical and physical concept, not just an AWS console dropdown. On the same day, Google locked in another 500MW of Texas solar (with a big battery attached), and Nordic banks wrote a $790m cheque for an AI campus in Narvik—because if uptime is the product, power and resilience are the inputs.

The Big Stories

AWS UAE outage prompts cloud region migration guidance is the one to sit with. After drone strikes damaged AWS data centres in its Middle East (UAE) region, AWS is advising customers to migrate workloads to other regions, with disruptions expected to last months. Beyond the immediate customer scramble, this lands as a real-world stress test of regional concentration risk: boards that treated “multi-AZ” as a complete answer are about to learn the difference between availability engineering and physical security.

Google signs 15-year PPA with Linea for 500 MW adds another chunky block of clean power to ERCOT. Google’s 15-year deal covers 500MW from the Duffy Solar Project in Texas, spanning 3,526 acres and co-located with a 235MWac battery energy storage system, with construction starting in Q3 2026. This is becoming a pattern rather than a one-off: the company is stacking gigawatts of PPAs (including a 1GW TotalEnergies deal and 1.2GW with Clearway) to de-risk power procurement for data centre growth in a market that can actually build.

Nordic lenders are getting comfortable treating GPU campuses like heavy industry. In Nordic banks fund $790M for Nscale AI data center, ABN A, DNB, Eksfin, Nordea and SEB provided $790m to Nscale for its Narvik AI campus, including an accordion facility tied to a potential 115MW expansion. It follows Nscale’s $2bn Series C and a $1.4bn delayed-draw term loan—capital structures that look less like “tech” and more like long-duration infrastructure backed by energy access.

Equinix is pushing deeper into Southeast Asia with a very explicit AI/HPC posture. Equinix to invest $190M building KL2 data center Kuala Lumpur will deliver more than 2,200 cabinets in Cyberjaya, integrated with Equinix Fabric, and designed to support advanced liquid cooling for Distributed AI and HPC workloads. The interesting bit isn’t just the capex number; it’s the signal that the premium colocation players are trying to stay “AI-relevant” in markets where power, land, and permitting can look very different from the US.

Meanwhile, grid reality keeps punching holes in everyone’s timeline assumptions. Downstream bottlenecks delay AI data center energization timelines points to PJM data showing AI infrastructure projects entering service in 2025 averaged more than seven years from interconnection request to operation—over three years to secure an Interconnection Service Agreement, then roughly four more years to energize. With transformer lead times “well over 100 weeks” and a projected equipment market growing from $20bn to $65bn by 2030 (per Wood Mackenzie, cited), the choke point has shifted downstream: you can raise money and pour concrete, but you can’t will transmission capacity and hardware into existence.

Behind the Headlines

Heat reuse is quietly becoming a differentiator, not a nice-to-have. Telehouse Canada deploys direct liquid cooling for AI workloads completed direct liquid-to-chip cooling across its Toronto campus, enabling up to 120kW per rack—and then did the part most operators still struggle to operationalise: capturing up to 80% of server heat and sending it into Enwave’s closed-loop district energy system to heat municipal drinking water. The takeaway isn’t the PUE/WUE talking points; it’s that interconnection-heavy urban campuses can turn “waste” heat into an asset if there’s a real offtaker and the plumbing is already there.

Ford’s move into stationary storage is a reminder that data centre power markets are now big enough to attract industrial-scale entrants. In Ford launches Ford Energy subsidiary and 20-foot BESS product, Ford created a wholly owned storage unit, outlined a 20-foot DC containerised BESS (5.45MWh using 512Ah LFP cells), and set a target of 20GWh of annual manufacturing capacity with first deliveries in late 2027—backed by about $2bn into a Kentucky production line. If you’re a data centre developer, the near-term impact is limited by that timeline; the strategic impact is bigger: the storage supply chain is being pulled toward “standardised, factory-built” products at scale, and that’s exactly what grid-constrained campuses will be buying.

Europe’s electrical equipment buildout is trying to catch up to demand that’s no longer hypothetical. ABB to invest $200m to expand European medium-voltage manufacturing lays out ~$200m over three years for medium-voltage capacity, including a $100m new facility in Dalmine, Italy, plus expansions across five other European sites—and an agreement to acquire Italian transformer maker Specialtrasfo (closing expected Q3 2026), which did ~€80m of revenue in 2025. Put alongside the transmission and transformer constraints now showing up in interconnection queues, this is what “AI buildout” looks like in the real economy: not just GPUs, but switchgear, transformers, and the factories to make them.