March 05, 2026

The most fragile part of “cloud resilience” isn’t software — it’s geography. Drone strikes damaged multiple AWS Availability Zones in the Gulf, taking two UAE zones offline and impacting one in Bahrain, and suddenly every board-level conversation about AI uptime has a new line item: what happens when a whole campus is a target.

That shock lands on the same day as a huge power-capital deal for the data-centre supply chain, and a White House-friendly pledge from Google that (quietly) tries to rewrite who pays for grid upgrades.

The Big Stories

AES is being taken private in a $33.4bn enterprise-value deal by a GIP- and EQT-led consortium that also includes CalPERS and Qatar’s sovereign wealth fund. AES isn’t shy about why capital matters here: it has 8.2GW of signed data-centre agreements and an 11.1GW renewables backlog, and the buyer group is explicitly framing this as fuel for U.S. generation, transmission, renewables, and data-centre supply. For investors, this is a tell that “power for compute” is graduating from a thematic slide to a buyout thesis — and it raises the bar for smaller IPPs trying to fund both renewables builds and the messy interconnection work behind them.

Google endorsed the White House Ratepayer Protection Pledge, committing to pay 100% of the power for its data centres, fund infrastructure costs driven by its growth, and bring net-new energy to the grid. Google points to adding more than 22GW of new energy globally over the past decade, and says it will use the Capacity Commitment Framework (CCF), pursue advanced nuclear, geothermal, and long-duration storage, explore a Clean Transition Tariff, and expand workforce training to lift the electrical workforce pipeline by 70% within five years. The subtext: hyperscalers can feel the political temperature rising around “who pays” — and they’d rather pre-commit than get boxed in by state-level moratoria, punitive tariffs, or blunt interconnection rules.

Nvidia will invest $2bn with Lumentum in advanced AI optical networking, pairing the money with multiyear supply agreements for optical components used in large AI systems. The collaboration spotlights silicon photonics and includes plans for a new U.S. fabrication facility led by Lumentum. This is Nvidia pulling a key bottleneck closer to home: once clusters scale, optics and interconnect become as existential as GPU supply, and “U.S. manufacturing capacity” is now a performance and geopolitical hedge, not just an industrial-policy talking point.

Pennsylvania is turning into a front line for data-centre permitting politics. Community groups, environmental activists, and bipartisan lawmakers are pushing back against more than 50 proposed facilities statewide, with a rezoning denial in Montour County and proposals for a three-year moratorium; legislators have also advanced HB 2151 to create a model municipal ordinance. The story explicitly ties local opposition to big-ticket power arrangements, including an AWS plan to pay $18bn to Talen Energy for up to 1.92GW of nuclear power through 2042. The warning for developers is simple: even “clean” power narratives don’t immunize projects from local land-use backlash — if anything, the scale now makes every project feel like a statewide issue.

The Gulf AWS disruption is the week’s clearest reminder that resiliency planning can’t stop at “multi-AZ.” With multiple Availability Zones knocked out in one area, cloud architects and risk teams are being pushed toward cross-region failover, a harder look at geopolitical risk, and even the uncomfortable idea of decentralizing compute away from concentrated hyperscaler campuses. If you’re selling colocation, network, or power into these ecosystems, this is the sort of event that accelerates spending — but not necessarily where incumbents expect it (think: connectivity, replication, and regional diversification rather than more density in the same metros).

Behind the Headlines

North American wind and solar PPA prices rose nearly 9% in 2025, according to LevelTen Energy’s January marketplace data, with ERCOT wind PPAs up 19% and late-2025 solar PPAs up 3.2%. The report links a late-2025 development surge to the One Big Beautiful Bill Act and rising data-centre demand, and it flags growing interest in energy service agreements, with median four-hour battery contract prices around $13/kW-month. The important nuance is that this isn’t just “renewables got pricier”: it’s the cost of certainty rising — for offtakers trying to lock in power, and for grids trying to absorb load growth while juggling interconnection and congestion. When the cheapest clean megawatt isn’t reliably deliverable, batteries and contract structures start to matter as much as the generation itself.

Capital Power’s 2025 results underscored how quickly power portfolios are being rebuilt around U.S. load growth. The company completed an approximately C$3.0bn acquisition of Hummel and Rolling Hills, adding ~2.2GW in PJM, reported full-year AFFO of C$1,066m, and disclosed MOUs including an Apollo-led investment partnership with up to US$3.0bn of potential committed equity. Tucked into that is a 250MW Alberta data-centre ESA expected to start in 2028 — a reminder that utilities and IPPs are increasingly structuring data-centre demand as contracted products (ESAs, capacity frameworks, bespoke tariffs) rather than just “new load.” For infrastructure investors, the financing and contracting innovation is becoming the competitive edge, not only the steel in the ground.

New South Wales endorsed AU$34.4bn of projects via its new Investment Delivery Authority, dominated by AU$34bn across 14 energy projects spanning pumped hydro, battery storage, wind, solar PV, and gas across multiple REZs (with developers including AGL, Neoen, Spark Renewables, Origin, and Santos). This isn’t a data-centre story on the surface — but it’s the kind of policy-backed throughput that determines whether Australia can keep approving high-density compute without the usual “grid can’t cope” handbrake. The tell is the mix: storage and firming are being pushed alongside new generation, implying the state is prioritizing system operability, not just nameplate capacity. If you’re planning large loads, you want to be in jurisdictions that can actually deliver multi-project coordination — because the alternative is death by queue position.