March 07, 2026

The most telling signal today wasn’t a new campus or a fresh funding round — it was politics catching up with physics. President Trump’s nonbinding “Rate Payer Protection Pledge” for data centers would push hyperscalers to bring on-site power and pay for grid upgrades, with Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI signing on. The catch is buried in the fine print: experts are already warning interconnection and permitting routinely run 5+ years, so “new power” simply won’t show up in time for 2027–2028 demand.

The Big Stories

The White House-backed pledge is a political attempt to stop data centers from becoming the villain in retail electricity bills. The pledge asks hyperscalers to supply on-site energy and cover grid upgrade costs — a clear nod to public frustration in constrained markets — but it’s explicitly nonbinding, and Sen. Mark Kelly dismissed it as a “handshake deal.” The real impact is directional: it normalises the idea that big AI buildouts must arrive with dedicated generation, which pulls the industry further toward behind-the-meter power, energy parks, and anything that shortens the queue for transmission and interconnection.

Europe got a very different kind of “power + compute” headline: Lyten’s acquisition of Northvolt’s Swedish battery assets, valued at nearly $5 billion, comes with plans for a Skellefteå “Industrial Hub” that pairs battery manufacturing with AI data centre development. EdgeConneX (EQT-owned) is also planning to acquire a Skellefteå data centre site that could scale to a one-gigawatt campus. This isn’t just another Nordic campus pitch — it’s a model investors should take seriously: colocate energy-intensive compute with industrial-scale energy infrastructure and manufacturing, then make the site expandable to GW-class.

On the demand side, Broadcom’s Hock Tan saying there’s a path to $100 billion of AI chips by 2027 is the kind of number that resets internal planning spreadsheets across the sector. Broadcom’s Q1 AI revenue more than doubled to $8.4 billion, and it forecasts $10.2 billion in AI chip revenue this quarter. The most consequential detail is the hyperscaler “capacity/commitments” it cites — Google 3 GW, Anthropic 3 GW, Meta at least 2 GW, OpenAI 1 GW — because that’s the power story leaking directly into the chip story. If those GW figures are even directionally right, the constraint isn’t “AI appetite,” it’s how quickly the physical stack (power, land, cooling, interconnect) can be brought online.

Capital is increasingly chasing solutions that bundle electrons with real estate. Schroders Greencoat’s new 50:50 JV platform with Greencoat Renewables is explicitly framed around “green energy data centres and energy parks” for hyperscalers, with an initial focus on Ireland. Its first acquisition is the Premier Periclase facility about 40 km north of Dublin, slated to become the Drogheda Energy Park with on-site generation, storage, advanced grid services, and corporate PPAs. Read that as a financial-product response to grid scarcity: if you can’t count on the utility timeline, you buy the ability to self-provide and trade grid services.

Networking power draw is creeping into the spotlight too. Cisco’s push into Linear Pluggable Optics (LPO) for Silicon One — including 800G LPO modules for Nexus 9000/8000 and its 102.4 Tbps G300 chip — comes with a bold claim: validated deployments can deliver 30%–50% power reductions. The fine print matters: LPO requires pairwise host–optic testing and careful reliability validation, which hints at friction in operations and supply chain qualification. Still, the direction is obvious: as AI clusters sprawl, shaving watts in the network stops being “nice-to-have” and becomes part of the capacity plan.

Behind the Headlines

The on-site power conversation is no longer theoretical, and manufacturers are tooling up accordingly. INNIO opening a new 6,400-square-meter plant in Hall in Tirol to expand Jenbacher engine production is framed directly around supplying data centers with decentralized energy solutions. Partial production began at the end of 2025 and full production has been operational since February 2026 — a timeline that reads like an industry sprint, not a cautious capacity add. What’s notable is how “decentralized energy” is being treated as a default architecture choice for data centres, not a temporary patch; once that mindset takes hold, it changes procurement, site selection, and even the competitive landscape for utilities.

Local permitting friction is still the quiet killer of deployment schedules, especially when projects touch water and noise. In Michigan, Allen Park’s Planning Commission delayed a decision on a 26 MW Solstice Data facility pending more noise and water-use studies; a planning study estimated over 3 million gallons of trucked water annually, 12 diesel generators, and generator noise up to 108 dB(A). The developer is pitching 200 temporary construction jobs, 30 permanent jobs, and $6.2M–$7.4M in annual property tax revenue — but the pause shows where community debate is heading: not “do we want digital infrastructure,” but “what exactly are you bringing with it?” Expect more projects to front-load detailed disclosures on water logistics, generator regimes, and mitigation plans if they want predictable timelines.

Southeast Asia is trying to write a more data-centre-friendly power story, but the storage piece is lagging. Thailand’s draft PDP 2024 and EGAT targets call for 35GW of renewables by 2037 (including 24GW of terrestrial solar) and 10GW/10.5GWh of battery storage plus 3.5GW pumped hydro, with EGAT also planning 2,725MW of floating solar across nine dams. Gulf Energy’s THB60 billion (US$1.9 billion) loan to invest in renewables and waste-to-energy, and a 2GW “DPP pilot” aimed at data centres requiring renewable power starting this year, show the region understands the buyer. But the gap between renewable build targets and a functioning storage market is where project risk hides — especially for AI loads that don’t tolerate supply volatility.