May 30, 2026

DigitalBridge’s move to buy ArcLight is the clearest signal yet that “AI infrastructure” is collapsing into a single trade: power plus digital, under one roof. In a deal valuing ArcLight at up to $1.05bn, DigitalBridge is explicitly trying to fuse power investing with data centres and connectivity—because the bottleneck isn’t racks, it’s electrons. Pair that with new US state scrutiny of AI data centres and fresh grid-rate manoeuvring in Texas, and the direction of travel is obvious: the next wave of winners will be the ones who can secure power, permits, and political permission at the same time.

The Big Stories

DigitalBridge to acquire ArcLight to scale power and AI in a transaction valuing ArcLight at up to $1.05bn ($650m base plus up to $400m contingent), creating a combined platform with more than $150bn of assets. The deal is also explicitly conditioned on SoftBank’s acquisition of DigitalBridge, and it’s framed as a way to align power and digital infrastructure investing for AI-driven demand. The important bit isn’t the headline valuation—it’s the admission that the data-centre buildout is now inseparable from generation, transmission, and the capital structures that fund them.

New Jersey proposes stricter oversight for AI data centers, with Governor Mikie Sherrill outlining a four-part plan that would force operators to fund infrastructure upgrades, disclose energy and water usage, and meet community standards. The politics here are sharpening: PJM grid strain gets cited, household bill impacts are put on the record (about $260), and the state points to roughly 80 data centres and an estimated $17bn contribution in 2023. If you’re underwriting US capacity growth, assume “data centres as a regulated load” is moving from fringe talking point to live legislative risk.

In Texas, TNMP files comprehensive base rate settlement with PUCT to recover a $2.8bn rate base, keep a 9.65% ROE and 45% equity ratio, and add a $20.5m Hurricane Beryl rider over five years—subject to PUCT approval. The notable twist is the coalition: PUCT Staff is joined by stakeholders including the Data Center Coalition, Walmart, and Amazon Data Systems. This is what the next phase looks like—large-load customers showing up early in rate cases because the cost of wires, storm hardening, and interconnection is becoming central to data-centre economics.

In Southeast Asia, DDSP secures $283M green loan for Johor hyperscale data centre to build a 45MW, AI-ready campus in Sedenak Tech Park, contracted to a global hyperscale customer. The financing syndicate (CTBC, MUFG, Standard Chartered, Entie) and the “green” label are doing real work here: capital is still available for scaled builds, but it increasingly wants a clear sustainability wrapper and an identifiable end-customer. Also worth clocking: DDSP is positioning this inside a much larger 1.1GW APAC pipeline—Johor isn’t a one-off, it’s a repeatable template.

The UK is moving to harden a different kind of critical infrastructure: UK to consult on tougher penalties for subsea cable damage, including tougher fines and prison sentences, plus new legislative proposals and a decision on a UK-flagged sovereign repair capability due by year-end. Government rhetoric is tying subsea resilience to “tens of billions of pounds” of private investment in UK AI infrastructure, and it even namechecks a £600m deal to unlock the Eastern Green Link 4 subsea energy project. Translation: cable security is being pulled into the same national-interest frame as power and AI compute—expect more regulation, more obligations, and potentially a new market for repair capacity.

Behind the Headlines

Water is quietly becoming the next gating item, and today’s datapoints put numbers on it. AI boom forces water constraints on data center expansion flags municipal water and wastewater capacity as an emerging hard constraint, citing a Virginia project asking for up to 2 MGD initially and up to 8 MGD in later phases, with explicit requirements for continuous evaporative cooling. The bigger punchline is the aggregate estimate: 697 million–1.45 billion gallons/day of new US water capacity through 2030, while Texas’ draft State Water Plan pegs roughly $174bn of water infrastructure projects over 50 years. Investors keep modelling “power + land + fibre”; they should be adding “water rights + wastewater capacity + local politics” as co-equal variables.

The UK’s “compute industrial policy” is getting more specific—and more hardware-shaped. Dell helps build UK’s Sunrise fusion AI supercomputer for research is a £45m government-backed system: 1.4MW, 6.76 exaflops of AI-accelerated modelling, targeted for June 2026, with Dell supplying the infrastructure and partners including AMD, Intel, WEKA, Cambridge, and UKAEA. This matters beyond the press release because it’s a blueprint for how governments are buying capability: not generic “AI funding,” but named workloads (fusion), measured power envelopes (MW), and identifiable supply-chain stacks. If you sell infrastructure, the buyer increasingly wants an outcomes story and a timetable, not just a cluster.

India continues to market itself as the cost-and-power counterweight to mature hyperscale markets. Telangana readies power and incentives for data centre boom says the state is prepared to supply clean, scalable energy for an anticipated 8–10GW of data-centre interest around Hyderabad, backed by the Telangana Clean Green Energy Policy 2025. Industry speakers cite a stark cost comparison—about $7m/MW in India versus $14m/MW in the US—while warning about skills shortages and urging green energy and batteries. The subtext: if India can reliably deliver electrons and talent at that price point, it won’t just absorb overflow demand; it could reset where the next round of capacity gets built.